US 9,185,095 B1Grant
Behavioral profiling method and system to authenticate a user
Issue Date:2015-11-10
•20 Claims
•8 Drawing Sheets
Abstract
Methods and systems for behavioral profiling are described. In some embodiments, a method includes observing behavioral characteristics of user interactions during a current session with the user through one of a plurality of channels. Variations between the behavioral characteristics of the user interactions observed during the current session and a behavioral profile previously developed based on prior usage patterns of the user through the plurality of channels are identified, in real-time or near real-time. For the user to proceed in the session, a challenge level is implemented based on the variations between the behavioral characteristics and the behavioral profile.
Metadata
Assignee
- United Services Automobile Association (USAA)
Inventors
- Karen Moritz
- Stephen Seyler Aultman
- Joseph James Albert Campbell
- Debra Casillas
- Jonathan Edward Neuse
- Sara Teresa Alonzo
- Thomas Bret Buckingham
- Gabriel Carlos Fernandez
- Maland Keith Mortensen
Application Information
Application Number:US 13/801,784
Filing Date:2013-03-13
Priority Date:2013-03-13
Art Unit:2493
Classifications
IPC:
H04L29/06
Field of Search:
G06F 21/00G06F 21/316G06F 3/0481G06F 3/0482G06F 3/016H04W 12/00H04L 63/00H04L 63/102H04L 63/08
Patent Drawings (8 sheets)
Description
Cross-Reference to Related Applications
[0001] This application is related to U.S. patent application Ser. No. 13/425,227, filed Mar. 20, 2012 and to U.S. Provisional Application Ser. No. 61/704,180 filed Sep. 21, 2012, each of which are hereby incorporated by reference in their entireties for all purposes.
Technical Field
[0002] Various embodiments of the present disclosure generally relate to authenticating a user. More specifically, various embodiments of the present disclosure relate to a behavioral profiling method and system to authenticate a user.
Background
[0003] Organizations strive to ensure secure and convenient user access to services or accounts. With the proliferation of identity theft and the growing emphasis on convenience, organizations are forced to find a balance between gathering enough identifying information to provide enough confidence in a user's identity and making the services or accounts accessible to users. Regulations and business rules may govern how much or what identifying information the user must provide depending upon the nature of the activity that is requested.
[0004] Many traditional systems often rely on authentication measure that include ownership factors (e.g., security token), knowledge factors (e.g., passwords or PINs), and/or inherence factors (e.g., fingerprints or retinal patterns). These authentication measures are often static or presented at pre-defined entry points. To varying degrees, authentication measures based on these factors can be easily circumvented and/or counterfeited. In addition, many systems simply rely on passwords and/or challenge responses at a single point to determine if access should be granted. Also, many systems sacrifice increased security measures for convenience of use. As such, techniques are needed that provide for improved security without sacrificing ease of use for the user.
Summary
[0005] In some embodiments, a method and system of user verification is described. A method may include observing behavioral characteristics of user interactions during a current session with a user through one of a plurality of channels; identifying, in real-time or near real-time, variations between the behavioral characteristics of the user interactions observed during the current session and a behavioral profile previously developed based on prior usage patterns of the user through the plurality of channels; and implementing a challenge level to proceed in the session, the challenge level based on the variations between the behavioral characteristics and the behavioral profile.
[0006] In some embodiments, the method further includes receiving current device information. Identifying variations between the behavioral characteristics may include comparing the current device information with historical device information stored in the behavioral profile. The current device information may include at least one of the following: device location, device identification, channel usage on the current device, language, network, or internet service provider. Identifying variations may include estimating a distance between behavioral characteristics in the current session and the behavioral profile.
[0007] In some embodiments, the method further includes developing the behavioral profile by identifying typical usage patterns of behavior from historical usage data; calculating a distance between the behavioral characteristics of the current session and the behavioral profile; and validating the behavioral profile during the current session when the behavioral characteristics of the current session are within a predetermined distance from the typical usage patterns of the user.
[0008] In some embodiments, the behavioral profile includes the behavioral characteristics and the method further includes representing at least a portion of the behavioral characteristics as metrics.
[0009] In some embodiments, the behavioral profile includes the behavioral characteristics, and wherein the characteristics include at least one of: duration between keyboard strokes, intensity of keyboard presses, user central processing unit time, system central processing using time, an amount of time between user logons, the amount of character IO during application execution, maximum size of a request, real user usage over time, virtual size of text segment, total number of ties opened during application execution, a number of major and minor page faults, a number of pages accessed, a measure of disk IO, elapsed time in seconds of the session, a number of signals received during application execution, a name of new user name changed during a session, a number of unique Internet protocol addresses used per month, hour of day of the session, number of distinct pages accessed during the session, whether an application executed was executed on a remote network host, a number of computers used each month, a name of a remote network host if an application was executed on the remote network host, name of a local network host on which an application was executed, number of different users with a same internet protocol addresses, number of seconds since a last audit record for an application, or a number of times the user logs in per predetermined time period.
[0010] In some embodiments, the method further comprises adapting the behavioral profile based on the behavioral characteristics of the user interactions observed during the current session. The behavioral profile may initially be created using demographic data of users similar to the user. The method may further include removing or deemphasizing at least a portion of the demographic data from the behavioral profile as the user behavioral profile is adapted with the behavioral characteristics of the user interactions observed during the current session.
[0011] In some embodiments, the method further includes receiving a response to the challenge level, determining whether the response to the challenge level authenticates the user; and adapting the behavioral profile based on the behavioral characteristics that triggered the challenge level. The challenge level may include allowing the user to proceed with the session, collecting identifying information, noting suspicious activity, or disallowing the user to proceed with the session.
[0012] In some embodiments, the variations are indicative of a second user, and the method may further include determining that the second user is authorized by the user, and developing a behavioral profile for the second user.
[0013] The challenge level may be based on a risk level of requested activities of the session and the behavioral profile may be based on authentication logs, clicktrail data, and previous warnings indicating suspicious activity. The plurality of channels may include an internet portal, face to face contact, a mobile application, and an instant messaging system.
[0014] The behavioral profile may be developed using at least one of the following: Bayesian network, statistical-based anomaly detection techniques, one or more Markov models, knowledge-based techniques, neural networks, clustering and outlier detection, demographic analysis, genetic algorithms, or fuzzy logic techniques.
[0015] In other embodiments, a computer-implemented method of fraud prediction including passively identifying a user interacting through a channel during a session; retrieving, from a database, a predictive behavioral profile associated with the user, wherein the behavioral profile receives current user interactions with the channel and estimates a distance from prior usage patterns of the user; identifying, using a processor, in real-time or near real-time, variations between current usage patterns of the user and the behavioral profile; and implementing a challenge level to proceed in the session, the challenge level based on the variations between the behavioral characteristics and the user behavioral profile.
[0016] The computer-implemented method may further include developing the predictive behavioral profile using at least one of the following: a Bayesian network, a statistical-based anomaly detection technique, one or more Markov models, knowledge-based techniques, neural networks, clustering and outlier detection, demographic analysis, genetic algorithms, or fuzzy logic techniques.
[0017] In other embodiments, a system for authenticating a user is disclosed. The system may include a channel communication module for engaging in one or more sessions with a user via a plurality of channels; an information gathering module for: monitoring user behavior during the one or more sessions and collecting demographic data relating to the user; means for developing a user behavioral profile based on the user behavior and demographic data, the user behavior profile including patterns of behavior that are typical of the user; the channel communication module for observing the user behavior during the session; a variation determining module for determining, in near real-time, variations between the user behavior observed during the session and the user behavioral profile; and a challenge module for implementing a challenge level to proceed with the session based on the variations; and adapting the user behavioral profile with the user behavior from the session.
[0018] While multiple embodiments are disclosed, still other embodiments will become apparent to those skilled in the art from the following detailed description, which shows and describes illustrative embodiments. As will be realized, embodiments of the present disclosure are capable of modifications in various aspects, all without departing from the scope of the present disclosure. Accordingly, the drawings and detailed description are to be regarded as illustrative in nature and not restrictive.
Brief Description of the Drawings
[0019] Embodiments of the present disclosure will be described and explained through the use of the accompanying drawings in which:
[0020] FIG. 1 illustrates an example of an operating environment in which a behavioral profiler may be utilized;
[0021] FIG. 2 is a block diagram illustrating components that can be included in a behavior profiler;
[0022] FIG. 3 illustrates a hypothetical deterministic finite state automation in accordance with various embodiments of the present disclosure;
[0023] FIG. 4 illustrates a hypothetical neural network model in accordance with various embodiments of the present disclosure;
[0024] FIG. 5 is a flowchart illustrating a set of operations for determining whether a user may engage in an activity in accordance with various embodiments of the present disclosure;
[0025] FIG. 6 is a flowchart illustrating a set of operations for verifying a user in accordance with various embodiments of the present disclosure; and
[0026] FIG. 7 illustrates an example of a computer system with which some embodiments of the present disclosure may be utilized.
[0027] The drawings have not necessarily been drawn to scale. For example, the dimensions of some of the elements in the figures may be expanded or reduced to help improve the understanding of the embodiments of the present disclosure. Similarly, some components and/or operations may be separated into different blocks or combined into a single block for the purposes of discussion of some of the embodiments of the present disclosure. Moreover, while embodiments of the present disclosure are amenable to various modifications and alternative forms, specific embodiments have been shown by way of example in the drawings and are described in detail below. The intention, however, is not to limit the disclosure to the particular embodiments described. On the contrary, the disclosure is intended to cover all modifications, equivalents, and alternatives falling within the scope of the disclosure.
Detailed Description
[0028] Various embodiments of the present disclosure generally relate to authenticating a user. More specifically, various embodiments of the present disclosure relate to a behavioral profiling method and system to authenticate a user.
[0029] Although this disclosure discusses the methods and systems for behavioral profiling in the context of fraud detection, one of ordinary skill in the art will recognize that the concepts disclosed herein may be used in various other areas such as marketing initiatives. For clarification, a behavioral profile means a history of past behaviors in combination with derived metrics that are created by applying mathematical and statistical functions to the history of past behaviors. Additionally, for clarification, this application will govern any inconsistencies with material incorporated by reference.
[0030] The increased sophistication and diversity of security attacks on electronic systems have forced companies to be reactive in fighting the resulting fraud. For example, a company may not be aware of a security attack that resulted in account takeover (“ATO”) fraud until an account owner reports it to the company. At this point, the company attempts to understand and reconstruct the techniques used by the criminals to create the security breach. This retroactive approach typically does not meet expectations of the customers. Instead, the customer expects the organizations to take the appropriate precautions to protect their data. In addition, the lack of comprehensive proactive protection plans to prevent cyber-fraud may result in dissatisfied customers and/or legal liability. Thus, a proactive and flexible approach for detecting suspicious activity in real-time or near real-time is needed.
[0031] Methods and systems for verifying a user proactively based on normal user patterns are described. In some embodiments, a behavioral profile can be created based on patterns of user behavior. As the user begins a new session interacting with the company (e.g., through a website, phone, or other channel), the interactions (or behaviors) are observed and constantly compared to normal patterns of interaction identified by the behavioral profile. A fraudster's behavior will deviate enough from a user's normal pattern that the presence of fraud can be detected before it affects the user.
[0032] Detecting these deviations, or anomalies, relies on creating a normal user profile by looking at the behavior of cross-channel contact the user has with an organization. The channels of interaction may include personal computers, mobile devices, face-to-face communications, and related hardware and software (e.g., web portals, mobile applications), and the like. The normal user profile created from the cross-channel contact could be based on online behavioral metrics such as the rate in which a user accesses the organization's digital channels, the number of pages he/she usually visits, the browser and device used to connect, and the number of different IP addresses used, etc. By comparing usage patterns in a current session to usage patterns in previous sessions, the previous usage patterns embodied in a behavioral profile reveals deviations.
[0033] Depending upon the importance or significance of the detected variations between the current usage patterns and the previous usage patterns, a challenge level may be implemented to proceed in the session. The challenge may require the user to provide additional information, allow the user to proceed with the session, or reject the user from continuing with the session. In some embodiments, the challenge is an authentication challenge. The user may be asked for additional information, but in some, embodiments, the user is authenticated without the user responding to a request (e.g., information is collected from the device). In many cases, the challenge level will be different based on the detected variation and the sensitivity of the information and/or activity being requested. For example, a request to see a balance could require a lower challenge level than transferring funds to a previously unknown account.
[0034] In the following description, for the purposes of explanation, numerous specific details are set forth in order to provide a thorough understanding of embodiments of the present disclosure. It will be apparent, however, to one skilled in the art that embodiments may be practiced without some of these specific details.
[0035] Having described embodiments of the invention generally, attention is now directed to FIG. 1 , which illustrates an example of an operating environment 100 in which some embodiments of the present disclosure may be utilized.
[0036] As illustrated in FIG. 1 , the operating environment can include dedicated terminal 110, ATM network 115, mobile device 120, personal computer 125, POS device 130, telephone 135, network 140, organization 145, behavioral profile database 150, security and fraud indicator database 155, behavior profiler 160, user database 165, and third party 170. Other operating environments may include different combinations of these and/or other devices.
[0037] The embodiments illustrated in FIG. 1 allow user interactions with organization 145 through various points of interaction such as, but not limited to, dedicated terminals 110 (e.g., public terminal or kiosk), ATM networks 115 (e.g., interbank ATM network and intrabank ATM networks), mobile devices 120 (e.g., mobile phone), personal computers 125, point of sale (POS) devices 130, by a landline telephone represented by telephone 135, and face-to-face contact (e.g., in-person contact or via video conferencing). The points of interaction may run one or more applications or clients that allow a user to interact with network 140. Such applications may provide access to the processing systems and interfaces of organization 145.
[0038] The points of interaction may be channels for a user to engage in activities with organization 145. For example, a user may use a web portal on a personal computer 125 (i.e. a channel) to engage in activities such as viewing an account, trading funds, purchasing insurance, or submitting or processing a negotiable instrument. Points of interaction telephone 135, POS device 130, mobile device 120, dedicated terminal 110, and their related hardware and software may be channels allowing for other user interactions such as video conferencing, interaction through a mobile application, or SMS messaging.
[0039] Dedicated terminals 110 such as public terminals/kiosks may be computers available to the public and/or specially designed kiosks that interact with network 140. ATM networks 115 can be any set of ATMs in an interbank ATM network and/or intrabank ATM network.
[0040] Mobile device 120 may be cellular phones, smart phones (a smart phone being a cellular phone that has capabilities and features such as, but not limited to, internet access, a full keyboard, email access, Wi-Fi connection capabilities, BLUETOOTH connectivity, or any other functionality of a computer), tablet computers (a portable computer having a touch interface), netbooks, laptops possibly with a wireless connection (such as an 802.11a/b/g connection or mobile broadband connection, such as via a 3G or 4G wireless network).
[0041] Personal computers 125 may be any computer (e.g., desktop computers, laptop computers, netbooks, tablet computers, Internet-enabled television devices, etc.) connected to network 140. Dedicated terminal 110, mobile device 120, and personal computer 125 may be capable of exchanging communication in the form of voice and/or video.
[0042] POS devices 130 can be any device used as a checkout or payment receiving mechanism for a merchant or service provider such as a hospital, ambulance company, doctor, automobile repair shop, pharmacy, clothing store, and others. For example, POS device 130 can be a terminal located at a merchant, a computer-based interface such as a webpage or custom application, a telephone payment system, and others.
[0043] Network 140 may include communications networks, such as a banking network. In accordance with embodiments of the present invention, network 140 can use a variety of interaction methods, protocols, and systems. For example, network 140 can use any of the automated clearing house (ACH) networks. An ACH network may be operated by NACHA (previously referred to as the National Automated Clearing House Association). Another ACH network may be the Electronic Payments Network (EPN). These ACH networks may interact to settle ACH transactions involving a party that has a relationship with only NACHA's ACH network or only the EPN. Other banking networks, such as CIRRUS, NYCE, and PULSE may also be used.
[0044] The contact with organization 145 may occur through sessions (e.g., to interactions devoted to a period of contact through a channel) at different points of interaction. Organization 145 may be a membership organization, and the users may be members or potential members of the membership organization. Organization 145 may be a financial institution, or any kind of entity providing any type of service.
[0045] Organization 145 may be communicably coupled with third party 170. Third party 170 is any organization or source of information external to organization 145 that provides organization 145 with information that may be used in determining potential or real threats. For example, information provided by third party 170 may include the identity of IP addresses known to be fraudulent, known mal-ware on systems or networks, patterns that indicate fraud, etc. Third party 170 may also provide information that is specific to a user. For example, third party 170 may provide information that a credit card belonging to the user has recently been compromised. The information may be stored in security and fraud indicator database 155.
[0046] Organization 145 may include various computing systems, behavior profiler 160, behavioral profile database 150, security and fraud indicator database 155, and user database 165. Behavior profiler 160 can use information from various databases and third party 170, including a behavioral profile related to the user, information related to the risk of the activity and information acquired in connection with the user, and determines a challenge level (e.g., an authentication level) that the user must surpass in order to continue with the session or particular activity within the session.
[0047] A session may be an interactive exchange defined by a beginning interaction, such as logging into an account, and an ending interaction such as logging out. For example, a session may begin when the user navigates to a webpage (i.e., the channel) and then end when the user navigates away. Similarly, a session may begin when a user initiates a phone call (i.e., the channel) with a customer representative and then end when the phone call is terminated. The interactions generated during each session can be monitored and/or recorded. Based on the interaction generated during sessions with organization 145, behavior profiler 160 dynamically can generate a current usage pattern for the session as information becomes available.
[0048] Behavior profiler 160 can be communicably coupled to one or more databases such as behavioral profile database 150, security and fraud indicator database 155, and user database 165, and may receive information from third party 170. These databases and external sources include information that may be used by behavior profiler 160. For example, behavioral profile database 150 may include user-specific behavior profiles generated to document normal usage patterns for the user. These profiles may be generated based on transaction information relating to past transactions such as the time of day transactions were made, the amount and destination of the transfer, the channels used, activities and times associated with those activities (time spent making the transaction), behavioral biometrics, etc. The profiles in the behavioral profile database 150 may also be based on information relating to the user's account, such as how many logon attempts have been made, the number of failed logon attempts, the number of successful logon attempts, where the attempts originated from, when/where/from what machine the user has changed passwords, registered devices, and so on. The profiles in the behavioral profile database 150 may be further based on the user's common purchasing habits.
[0049] User database 165 stores information about users, including employment information, account balances, credit ratings, home ownership information, annual salary, length of membership, and/or other information such as the information used in creating the behavioral profiles stored in behavioral profile database 150. User database 165 may store security preferences or instructions that the user has expressed (e.g., if a requested transaction exceeds $100, ask the requestor to verify a one-time password sent to an email address of the user).
[0050] Security and fraud indicator database 155 stores information relating to physical security events, information security events, cyber intelligence, IP addresses that organization 145 has previously received attacks from, addresses of known fraudsters, known mal-ware on systems or networks, risk scores associated with IP addresses or networks, etc. This information may also be provided or supplemented by third party 170. This information may be applicable to all users, or it may be related to a particular user noted in the user database 165 or included in the behavioral profile of the user stored in the behavioral profile database 150. Thus, this information may impact the riskiness of an activity for a particular user or impact the users globally. In some embodiments, these databases may be integrated into one database.
[0051] A computer system associated with the organization or authorized third parties may be able to access these databases for account information, user information, security information, and other stored information. In addition, third parties may provide information relating to the user such as the location of the user, a credit score of the user, or any other security information.
[0052] FIG. 2 is a block diagram illustrating components that can be included in a system such as behavior profiler 160. According to the embodiments shown in FIG. 2 , memory 210, processor(s) 220, channel communication module 225, activity request receiver 230, information receiving/extracting module 235, behavioral biometrics module 240, behavioral profile generation module 245, variation determination module 250, distance threshold module 255, challenge level module 260, validation module 265, response module 270, behavioral profile adaptation module 275, compliance module 280, and Graphic User Interface (GUI) generation module 285. Other embodiments of the present disclosure may include some, all, or none of these modules and components along with other modules, applications, and/or components. The modules and components illustrated in FIG. 2 may be hardware, firmware, or software or a combination. Still yet, some embodiments may incorporate two or more of these modules into a single module and/or associate a portion of the functionality of one or more of these modules with a different module.
[0053] For example, in one embodiment, the functionality associated with behavioral profile adaptation module 275 and the functionality associated with behavioral profile generation module 245 can be incorporated into a single behavioral profiling module. In other embodiments, the challenge level module 260 can be separated into several modules such as a challenge level determination module and a challenge level implementation module.
[0054] Memory 210 can be any device, mechanism, or populated data structure used for storing information. In accordance with some embodiments of the present disclosure, memory 210 can encompass any type of, but is not limited to, volatile memory, nonvolatile memory, and dynamic memory. For example, memory 210 can be random access memory, memory storage devices, optical memory devices, magnetic media, floppy disks, magnetic tapes, hard drives, SIMMs, SDRAM, DIMMs, RDRAM, DDR RAM, SODIMMS, erasable programmable read-only memories (EPROMs), electrically erasable programmable read-only memories (EEPROMs), compact discs, DVDs, and/or the like. In accordance with some embodiments, memory 210 may include one or more disk drives, flash drives, one or more databases, one or more tables, one or more files, local cache memories, processor cache memories, relational databases, flat databases, and/or the like. In addition, those of ordinary skill in the art will appreciate many additional devices and techniques for storing information which can be used as memory 210.
[0055] Memory 210 may be used to store instructions for running one or more applications or modules on processor(s) 220. For example, memory 210 could be used in one or more embodiments to house all or some of the instructions needed to execute the functionality of channel communication module 225, activity request receiver 230, information receiving/extracting module 235, behavioral biometrics module 240, behavioral profile generation module 245, variation determination module 250, distance threshold module 255, challenge level module 260, validation module 265, response module 270, behavioral profile adaptation module 275, compliance module 280, and GUI generation module 285.
[0056] Channel communication module 225 exchanges communications between the organization 145 and users through various channels. Examples of channels include, but are not limited to, computing devices such as tablets, personal computers, smartphones, POS devices, ATMs, and laptops; telephones such as landline telephones or a mobile phones; and face-to-face contact between a user and an employee at a brick and mortar building of the organization (e.g., data entry in the system in response to user interaction). Channels may also include software and firmware associated with the computing devices and communications devices (e.g., mobile phones) such as web portals, applications, networks, mobile applications, and instant messaging systems. Channels may also include hardware associated with the computing devices and telephones such as the network hardware, credit card scanners, and retinal scanners.
[0057] Users may interact with the organization via the channel communication module 225 during one or more sessions. As described above, a session may be an interactive exchange defined by a beginning interaction, such as logging into an account, and an ending interaction such as logging out. In some embodiments, sessions between the user and the organization may be contiguous time periods of interaction or related interactions that occur over a series of distinct time periods. In another example, a session may be a phone call, which begins with a user calling the organization and ends when the call is terminated. One or more activities may take place during a single session. For example, a website account session may begin with the user logging in using a username/password, completing several activities such as checking an account balance, ordering a new credit card, and transferring funds, and ending the session by logging out. A session generally takes place through a single channel. In some embodiments, however, multiple channels may be used in a single session by porting a session from one channel, such as a mobile device, onto a second channel, such as a computer.
[0058] During a session between the user and the organization, the user may request to engage in one or more activities. The request can be routed to the activity request receiver 230 once the request has been received. Channel communication module 225 may provide information relating to the channel, activity, and/or user to the information receiving/extracting module 235.
[0059] Activity request receiver 230 receives requests from a user to engage in an activity. Activity request receiver 230 receives the request through channel communication module 225 during a session with the user. An activity may be any activity in which a user interacts with organization 145, such as viewing a webpage, transferring funds, accessing an account, purchasing insurance, depositing funds, opening an account, paying a bill, and requesting a line of credit or a credit card. Due to regulatory and/or business restrictions, some activities may only be engaged in through specific channels.
[0060] Information receiving/extracting module 235 receives, collects, and/or extracts information relating to the user, activity, and/or channel. The information may include behavioral characteristics of the user interactions during a current session with the user. Such information may be used in the behavioral profile generation module 245, behavioral profile adaptation module 275, and/or variation determination module 250 in detecting fraud. For example, information receiving/extracting module 235 may receive current device information and provide it to the variation determination module 250 to determine if there are variations between the current device information and device information stored in the user's behavioral profile.
[0061] Information relating to the channel may include device information such as information relating to usage on the device, a device fingerprint or other device identification, a phone number, a service provider, device location information, language, network, internet service provider, internet protocol address, etc. Information relating to the channel may also include information relating to known risks, such as a risk of the location (e.g., IP addresses originating in China may be suspicious), risks associated with particular IP addresses or phone numbers, and the like. Information relating to the channel may be received from or collected by the channel communication module 225, securities and fraud indicators database 155, third party 170, and other sources.
[0062] Information relating to the activity may include a type of activity (e.g., deposit, transfer, logon request), a risk indication of the activity (e.g., transferring funds vs. depositing funds), and specific activity information such as time of day, amount of money, location of transfer, type of account, and reason for request. Information relating to the activity may be collected from activity request receiver 230, third party 170, user database 165, and other sources. Additional information may be received from channel communication module 225.
[0063] Information relating to the user may include web behaviors, security questions and answers, usernames, accounts, family information, websites visited, amounts, locations, devices, and activities. Information relating to the user may also include military status (active duty), military rank, marital status, gender, product depth and tenure with organization 145, and age of children (if any). In some embodiments, particularly when little information is known about the user, information receiving/extracting module 235 may collect demographic information data of others similar to the user. Demographic information may be collected from marketing groups or other sources.
[0064] Information receiving/extracting module 235 may also collect information relating to suspicious activities noted in the user's account. For example, a notice documenting an agent's impressions or concerns may be placed in the user's account indicating past fraudulent activity or suspicious activity. Information relating to the user may be collected from third parties 170, user database 165, and other sources.
[0065] Information specific to a current session with the user through a web portal (web behaviors) may be extracted, collected, or gathered such as: user CPU time, system CPU time, the amount of time between when a user logs in (when during the day/week/month does the user log in), amount of character IO during the application execution, combined process size—that is, the maximum size of the process during execution, the integral of real memory usage over time, in units of KByte/seconds, the virtual size of the text segment, the total number of ties opened during execution, the number of major and minor page faults, the number of pages read in, a measure of disk 10, elapsed time in seconds—that is, the exit time minus the start time of the application, number of signals received during the execution of the application, name of new user name (if the user name was changed during execution), the number of unique IP addresses used per month, whether the user name was changed during execution, hour (0-23) in which activity was done, the number of distinct pages a user normally views per session, whether the application executed was on a remote network host, the number of different computers used per month, name of remote network host on which the application was invoked, name of application that was invoked on a remote network host, whether the application executed was on a local network host, name of local network host on which this application was invoked, name of application that was invoked on a local network host, the number of different users with the same IP, number of seconds elapsed since the last audit record for this application, and the number of times a user logins per day/week/month. Other web behaviors may also be collected in some embodiments.
[0066] Behavioral biometrics module 240 can gather and analyze behavioral biometrics. Biometrics may be used to verify a user based on a biological attribute that is difficult to replicate. While traditional biometric approaches such as fingerprinting and/or retinal scanners work effectively, these approaches require specialized equipment. Behavioral biometrics identify behaviors that are heavily influenced by a person's specific biology but are easier to measure than the underlying biology itself. For example, users may be identified by measuring the cadence with which the user typed their username and password, spacing between key presses, and/or the duration of each key press. Markov models may be trained to model these and other behavioral biometrics of a user.
[0067] Information collected and analyzed by behavioral biometrics module 240 may be used in distinguishing multiple users of a single account. For instance, a user and the user's spouse may use the same credentials to logon and manage their checking and auto products. By using the keystroke-based biometrics, the number of different users of a single account may be identified, and behaviors may be more accurately attributed to different individuals.
[0068] Behavioral biometrics module 240 may analyze the user's behavioral biometric data and provide the user's behavioral biometrics information to the behavioral profile generation module 245 and/or the variation determination module 250 to provide additional information about the user and reduce false positive rates when detecting fraud.
[0069] Behavioral profile generation module 245 may generate behavioral profiles for each user. In general, a behavioral profile is developed based on prior usage patterns of the user through the interactions occurring through the various channels. The behavioral profiles may be generated in a variety of ways using a variety of methods and models which are described in more detail below.
Information Used in Generating Behavioral Profiles
[0070] A behavioral profile may include characteristics (e.g., behaviors, information) of the user. Typical usage patterns of behavior that the user may be identified from using historical usage data based on information may include: authentication logs, clicktrail data (e.g., online activity), user across channel business process history, user level demographic data (possibly from a marketing department associated with organization 145), previous warnings indicating suspicious activity, a list of high risk web pages (e.g., webpages associated with transferring money out of an account), and other information provided by information receiving/extracting module 235. These data sources may be used to create the behavioral profile in the following manners.
[0071] Authentication log data may track all attempted authentications with organization 145 as well as other data related to the authentication process. For instance, authentication logs include the date and time in which a user's password or security questions are changed. The authentication logs store authentication attempts across channels. Metrics such as the method and frequency with which the user changes his password (e.g., whether the user normally calls in to have their password changed) and how many times the user normally has to be authenticated may be used in building the user behavioral profile. This data can also be helpful in combination with all the other sources to help capture a normal path for users and fraudsters. For example, it could be out of the norm for a user to fail their online authentication forcing them to call in and have a representative update their information, but this could be a normal pattern for a fraudster.
[0072] The clicktrail data captures the activity of online users as they carry out their day-to-day transactions through the organization's website. The advantage of using clicktrail is that the sheer volume of transactions logs for each user makes it possible to develop a unique understanding of each user's website usage. Information as to which pages users visit, the order user visit pages, and how long users stay on each page is available within clicktrail.
[0073] The user across-channel business process history includes records the transactions or business processes that occur on each user's account across all channels. As the name implies, this data captures the transactions regardless of the channel on which a transaction is conducted. The across-channel understanding aids in the potentially multi-channel nature of ATO attempts and other fraudulent behavior. In an example of across-channel behavioral characteristics, each time a user deposits money in the user's account at a specific ATM (first channel), within 20 minutes, the user checks her account online to ensure that the funds were deposited. In another example, if a user typically requests a password reset via an email, a password request via phone call may indicate fraud. This data is also useful for establishing what transaction cadence is normal for each user across all of the channels and entities related to a company or organization.
[0074] Demographic data may be used in establishing behavioral profiles for users with relatively little history and may be used to establish conditional thresholds. Organizations may not initially have access to the type of specific information related to a user to build a behavioral profile. In such situations, a behavioral profile may initially be generated using demographic data of others that are similar to the user on the basis that what constitutes “normal behavior” for a user depends both on their own personal history as well as their life circumstances. Individuals with similar demographic profiles are more likely to have similar web behaviors. For instance, the threshold for investigating an unusual string of web transactions will likely be different for users in the baby boomers generation than for users in the millennial generation. In another example, a young, enlisted person who only has an auto policy with the organization may have different web behaviors than a married mother of four children who has all insurance and banking with the organization.
[0075] Demographic groups can be formed based on demographic categories such as: whether a user is active duty, military rank, marital status, gender, product depth and tenure with the company, and age of children. The groups can be formed on a more granular level by using these demographic categories to group users. By creating groups based on demographics and similar web behaviors, each user can have a group that represents that user without storing as much data as creating individual profiles for each person. In some embodiments, groups are defined by both demographics and the particular user's behaviors including web behaviors. In some embodiments, behavior profiles for each user include some demographic information relating to the user.
[0076] In some embodiments, the behavioral profile is initially created using demographic data, but the behavioral profile is adapted by the behavioral profile adaptation module 275 after data collection through one or more sessions with the user. Thus, the user's profile may be updated to include more specific user behaviors instead of demographic data.
Models
[0077] The behavioral profile may be developed using one or a combination of different models or techniques. For example, the behavior profile may be developed using at least one of the following: Bayesian network, statistical-based anomaly detection techniques, one or more Markov models, knowledge-based techniques, neural networks, clustering and outlier detection, demographic analysis, genetic algorithms, or fuzzy logic techniques. The below examples are described specifically in modeling web behaviors, but these techniques and models may be used in modeling behaviors in other channels, or a combination of channels.
Statistical-Based Techniques
[0078] In statistical-based techniques, web activity is captured and a profile representing its stochastic behavior is created. Early approaches to statistical-based anomaly detection systems were related to univariate models. Parameters were modeled as independent Gaussian random variables thus defining an acceptable range of values for every behavior variable. Group comparisons may be used to detect irregular patterns of spending behavior by observing individuals that began to behave differently from their established pattern. “Three-level-profiling” focuses on behavioral changes for the purpose of fraud detection. Three-level-profiling functions at the account level and indicates significant deviations from an account's normal behavior as a potential fraud.
[0079] Later, multivariate models that consider the correlations between two or more metrics were proposed. A multivariate weighting system may be applied to data to give older information less importance than new information in anomaly detection systems. Time series models may use an interval timer, together with an event counter or resource measure, and take into account the order and the inter-arrival times of the observations as well as their values. Thus, an observed traffic instance will be labeled as abnormal if its probability of occurrence is too low at a given time.
[0080] In statistical-based anomaly detection systems, no prerequisite for prior knowledge of normal activity exists because statistical-based systems can learn the expected behavior by simply observing behaviors. Furthermore, as periods of observation grow longer and longer, statistical methods provide increased accuracy in detecting fraudulent activity.
[0081] In some embodiments, outliers may be detected by using a multiplier for a standard deviate. In other embodiments, more recent observations are weighted more than observations used in the distant past.
[0082] One proposed method under statistical based anomaly detection is to create confidence sets or confidence intervals for univariate models. Suppose x1, . . . , xn are the observed behaviors of a user from clicktrail and e is the normal behavior parameter that is trying to be defined, then the interval estimate is a pair of functions L(x1, . . . , xn) and U(x1, . . . , xn) such that the inference L(x)≦θ≦U(x) is made. Parameters may be modeled as independent Gaussian random variables, thus, defining an acceptable range of values for every variable.
[0083] Another approach is to use an outlier detection method for univariate analysis which involves calculating a standard deviate for the parameter, θ, in order to define an interval. In any event, interval estimates of the behavioral metrics will be calculated so that coverage probabilities will cover the actual normal behaviors of each individual user with a high degree of certitude.
[0084] Another statistical-based technique proposed is multivariate models. In the multivariate case, Xi is defined as (Xi1; Xi2; . . . ; Xip)′, a vector of p measures on a behavioral process at time i. When the behaviors are normal, the population of X is assumed to follow a multivariate normal distribution with the mean vector μ and variance-covariance matrix Σ. Using a sample of data of size n, the sample mean vector X− and the sample covariance matrix S are usually used to estimate μ and Σ, where X =( X 1, X 2, . . . X p)′
and
[0085]
[0086] Hotelling's T2 statistic for an observation Xi can be determined by:
T2=(Xi−X )′S−1(Xi− X )
T2=(Xi−
[0087] A large value of T2 indicates a large deviation of the observation Xi from the observed normal pattern. The main difference between these and the univariate intervals is that multivariate models consider the correlations between two or more metrics. These are useful because experimental data have shown that a better level of discrimination can be obtained from combinations of related measures rather than individually.
Knowledge-Based Techniques
[0088] The expert system approach is one of the most widely used knowledge-based anomaly detection techniques. Electronic transactions may be procured and analyzed to reconstruct or detect a claim of fraud. Human pattern recognition skills may be combined with automated data algorithms. Information may be presented visually by domain-specific interfaces; that is, an automated algorithm filters which instances of fraud should be reviewed by expert auditors. One example specifies computer penetrations as sequences of actions that cause transitions in the security state of a system.
[0089] Forensic analytics may be used to review the invoicing activity for a vendor to identify fictitious vendors, and these techniques might also be used by a franchisor to detect fraudulent or erroneous sales reports by the franchisee in a franchising environment. Expert systems are intended to classify audit data according to a set of rules involving multiple steps: data collection, data preparation, data analysis, and reporting.
[0090] Methods where the desired model is manually constructed by a human expert could be potentially more restrictive. As long as the constructed specifications are complete enough, the model will be able to detect illegitimate behavioral patterns. In fact, the number of false positives can be reduced since this kind of system avoids the problem of harmless activities, not previously observed, being reported as intrusions.
[0091] Specifications have also been developed by more formal tools. Link analysis may be used to relate known fraudsters to other individuals by using record linkage and social network methods.
[0092] The most significant advantage of knowledge-based approaches to anomaly detection is their robustness and flexibility. In some embodiments, developing quality knowledge may be time-consuming.
[0093] An N-gram is a term from the field of computational linguistics and refers to a sequence of n items from a larger sequence. While N-grams are often used to look at the order of words in linguistic applications (i.e., sentiment analysis of tweets), they can also be used to analyze the sequence of webpages viewed by users. For instance, in one approach, the most common pages viewed preceding a “high risk” page is documented. By developing a baseline for each user of the most common N pages preceding the viewing of each high risk page, it is possible for an organization to detect abnormal web traffic over an arbitrary sequence depth to high risk pages.
[0094] In addition to calculating the most common routes to high-risk pages by the users, a similar calculation on web sessions known to be associated with account takeovers can be performed. This allows for a comparison between the most frequent routes to high risk pages for non-fraudulent traffic and the most frequent routes for malicious web traffic. The results of these analyses can, again, be setup as a system of flags. If a user takes a common route to a high risk page then that is a positive signal whereas if they take a route commonly used by malicious agents, then the web session could be flagged.
[0095] An advantage of the N-gram approach is that it can be expressed in terms of deterministic finite state. For example, the finite state machine methodology—a sequence of states and transitions among them—seems appropriate for modeling web behavior. Deterministic finite state automatons (DFAs) are a common abstraction used to describe high performance pattern matching algorithms such as regular expressions.
[0096] To help illustrate this idea, consider the following example. Suppose that two high risk navigating patterns have been observed in the data. The first pattern is A→B→C. The second pattern is B→D→E. In this example, pages C and E are high-risk web pages where it is possible for the association to be defrauded. From a technological perspective, it is desired to find a way to keep track of both navigational sequences that as pages are requested through the web sphere layer and to send a message/raise a flag in the event that one of the monitored sequences occurs. In some embodiments, this tracking method may be lightweight and have a small computational footprint. FIG. 3 displays a hypothetical DFA given the two example sequences, (ABC)|(BDE).
[0097] The circles with the indication “State” in FIG. 3 represent a finite set of states that occur. The arrows originating from a state show the conditions on which the machine should transition from one state to another. For instance, if the machine is in state 3 and the next page that a user requests is the page labeled D, then the machine should transition to state 4. Conversely, if the next page that is requested is page C, then the machine should raise a flag. For the sake of brevity, a number of transitions arrows have been omitted. The semantic meaning of the omitted arrows can be summed up by saying that if the next page that is requested does not have a corresponding arrow for that state, the machine should return to the “start” state. When implementing these DFAs in a programming language, “else conditions” may be used.
[0098] Performance of an implemented fraud detection model may be considered in constructing the behavioral profile model. The speed with which a web page loads has important ramifications on user perceptions of quality and enjoyment and may be tracked by search engines when calculating page rankings. When implementing a fraud prevention model like the one described here, most of the scoring can be asynchronous. In the example above, only pages C and E were high risk and thus it may only be necessary to check the model score when these two pages are requested. Model score can include many models providing input into decision-making. One architectural approach may be to use event processing to keep a running score of each session asynchronously, and to only request the model score (synchronous request) when a high risk page is requested.
Machine Learning Techniques
[0099] Machine learning techniques generally establish a model that analyzes patterns in data in order to classify an outcome. For example, machine learning techniques can analyze the patterns of normal web behaviors for users to determine whether or not fraud has occurred. Machine learning techniques require identified data to train the behavioral model; that is, a data set containing both fraudulent and non-fraudulent web sessions are required. The more accurate the data, the more predictive power a model. Applying machine learning techniques to anomaly-based detection is similar to applying statistical-based systems, though machine learning systems focus on a model that improves its performance by learning from previous results.
[0100] Consequently, a machine learning anomaly-based detection system can change its execution strategy as new information is obtained making it desirable for anomaly detection applications. In some embodiments, machine learning techniques have large resource requirements.
(i) Bayesian Networks
[0101] A Bayesian network is a model that deciphers the probabilistic relationships among variables. Several advantages, like incorporating both prior knowledge and data, come from combining this technique with other statistical schemes. Bayesian networks may be implemented to detect credit card fraud, telecommunications fraud, medical insurance fraud, and auto claim fraud. In some embodiments, a hybrid knowledge/statistical-based system where expert knowledge is integrated with statistical inference may be implemented.
[0102] The results from Bayesian networks are remarkably similar to those derived from less computationally intensive threshold-based systems. In some embodiments, Bayesian networks may be dependent on the assumptions about the behavior of the target system.
(ii) Markov Models
[0103] There are two main approaches to Markov models. First, a Markov chain is a set of states that are interconnected through certain transition probabilities, which determine the topology of the underlying process. Second, a hidden Markov model assumes that observed behaviors are a Markov process in which states and transitions are hidden. Only the “productions” are observable.
[0104] Markov-based techniques may be used in intrusion detection systems. Hidden Markov models may score the probability of observing sequences of behaviors against a baseline. In network intrusion detection systems, the inspection of packets has led to the use of Markov models in some other approaches as well. The derived model may provide a good approach for establishing a normal profile. In some embodiments, results may be dependent on the assumptions about the behavior accepted for the system.
[0105] Aside from being more computationally intensive than statistical-based techniques, hidden Markov models typically involve analysis from any one current state to the next succeeding state. While it is possible to conduct Markov chain analysis where the starting state is described in terms of the previous N states, doing so greatly increases the dimensionality of the problem.
[0106] The basis for Markov chain analysis is that there are a finite number of “states” that can be observed. In an example, each webpage is represented by a state. The next step in Markov chain analysis is to calculate the probability of starting in a particular state and then transitioning to all of the other possible states. This set of conditional probabilities is often expressed in table form. When represented in table form, it is often referred to as a transition matrix because the table displays the probability of transitioning from any pair of states.
[0107] For example, suppose the following sequence of letters is observed: a, b, b, a, c, c, a, c, c. Table 1 outlines the transition matrix with each cell representing the associated probability of transitioning from one letter to the next. Since the sequence a,a, is never observed, the associated probability is zero. Likewise, since the sequences b,c or c,b are never observed, their associated transition probabilities are also zero. For the transitions that are observed, this method calculates the associated probabilities of observing each transition. For example, the probability of observing the sequence a,c is two-thirds.
[0108]
| TABLE 1 | ||||
|---|---|---|---|---|
| Hypothetical Transition Matrix | ||||
| Ending | ||||
| Starting | a | b | c | |
| a | 0.00 | 0.33 | 0.67 | |
| b | 0.50 | 0.50 | 0.00 | |
| c | 0.33 | 0.00 | 0.67 | |
[0109] The application of Markov chains in this example is to exchange the letters in the example above with actual page stems from clicktrail. Using this mathematical framework, a baseline probability may be first calculated for each user. Once baseline transition behavior is established, it is possible to calculate the likelihood of observing any sequence of page views and determine if an observed transition is outside of what is expected. If the organization observes a web session that that has a very low probability of occurring based on past experience, the session could be flagged as abnormal. In addition, the number of low probability transitions will be counted, providing insight into abnormal behavior with less computational burden.
(iii) Neural Networks
[0110] Neural networks are useful for anomaly detection because of their flexibility. Neural networks are sophisticated modeling techniques that are able to model complex functions. They often boast more predictive power than more traditional predictive models since they can fit adaptable nonlinear curves to data. Neural networks are often used when the exact nature of the relationship between inputs and output is unknown. This detection approach may be used to create user profiles, to predict the next command from a sequence of previous ones, and to identify the intrusive behavior of traffic patterns.
[0111] In some embodiments, neural networks may require a large diversity of training for real-world operation. In other embodiments, neural network models may not provide a descriptive model that explains why a particular detection decision has been taken; that is, there would be no way to tell why a particular session was flagged as fraudulent. Identifying which behaviors or behavioral changes contributed most to labeling a web session as fraudulent may be useful, especially if a particular behavior is indicative of future attacks.
[0112] Neural networks come in all shapes and forms. Selecting an appropriate network structure helps in building a good predictive model. A general form of an example model is shown in FIG. 4 . Generally, neural networks are composed of an input layer, with as many nodes as the number of input behaviors and features being considered, and an output layer, which in this example is a regression function made up of a single node representing whether or not a web session is fraudulent. In between the input and output layers though, the neural network may be configured with any number of hidden-layers and nodes.
[0113] Choosing the correct number of nodes may be challenging. If too few hidden nodes for a neural network model are chosen, the mapping function between the web behaviors and the fraud indicator may not be learned. If too many nodes are chosen, the model over fits the data, making it unable to predict future events.
[0114] Hierarchical clustering methods offer n possible solutions for a data set with n objects. As such, these clustering procedures are time consuming. Moreover, in some embodiments, once the clusters are assigned, they cannot be amended. A useful alternative to hierarchical clustering techniques is the k-means approach. The most common k-means algorithm uses an iterative refinement technique. Given an initial set of k means: m1(t), . . . , mk(t) (see below), the algorithm proceeds by alternating between two steps:
[0115] 1. Assignment step: Each observation is clustered to the nearest clusters mean.
[0116]
where each observation, x_p, is assigned to exactly one cluster, S(t), even if it could be is assigned to two or more of them.
where each observation, x_p, is assigned to exactly one cluster, S(t), even if it could be is assigned to two or more of them.
[0117] 2. Update step: New means for each cluster are calculated as the centers of the observations within the new clusters. The algorithm has converged when the assignments no longer change.
[0118]
(iv) Fuzzy Logic Techniques
(iv) Fuzzy Logic Techniques
[0119] Fuzzy logic based techniques can be used in anomaly detection because many of the features considered in anomaly detection systems can be thought of as approximate rather than precise. Similar to univariate statistical based anomaly detection techniques, fuzzy techniques consider an observation as normal if it lies within a given interval. In some embodiments, high resource consumption may be used with fuzzy logic techniques.
(v) Genetic Algorithms
[0120] Inspired by evolutionary biology, genetic algorithms borrow from concepts like inheritance, mutation, selection, and recombination to deriving classification rules and/or selecting appropriate features or optimal parameters for the detection process.
[0121] One advantage of this type of machine learning anomaly detection technique is the use of a flexible and robust search method that converges to a solution from multiple directions. In some embodiments, no prior knowledge about behaviors is assumed. Its high resource requirement makes it less desirable compared to other methods.
(vi) Clustering and Outlier Detection
[0122] The observed user behavior may be aggregated, and the resulting aggregations could elucidate fraudulent behavior through group characteristics. Clustering techniques work by grouping the observed data into clusters using a similarity or distance measure. In general, a cluster begins with a representative point. Then, each new data point is considered and classified as either belonging to an established cluster or not according to the proximity of the corresponding representative point. Outliers are then points that may not belong to any cluster and could represent the anomalies in the detection process.
[0123] Determining whether or not an outlier is actually an anomaly is a heavily researched question among fraud researchers. Often, different clustering algorithms indicate different clusters, and within the fraud literature there is no real agreement on which algorithm works best. For example, some systems use the k-nearest neighbor approach with the Euclidean distance to define the membership of data points to a given cluster. Other systems use the Mahalanobis distance. These distance definitions can then be used to associate a certain degree of being an outlier for each point. Nevertheless, clustering techniques determine the occurrence of intrusion events using only raw audit data, and so the effort required to tune the anomaly detection system is reduced.
[0124] Variation determination module 250 can identify, in real-time or near real-time, variations between the behavioral characteristics of the user interactions observed during the current session and the behavioral profile previously developed based on prior usage patterns of the user through the channels or demographic information. Variations may be detected using the methods described above with respect to the particular methods or techniques of modeling behavioral profiles. For example, variations may include variations in behavioral biometrics (e.g., the username/password was typed at a different cadence than the user typically types the username/password), webpage viewing, device ownership, or any of the other behavioral characteristics discussed above in connection with the information receiving/extracting module 235.
[0125] In some embodiments, the variation determination module 250 identifies that a second user is using the account based, on the variations. Variation determination module 250 may further determine that the second user is authorized by the user by viewing a profile of the user or by other mechanisms (e.g., asking questions, alerting the user). For example the user may be the spouse of the user and the two may share the account. Behavioral profile generation module 245 may then generate a behavioral profile for the second user.
[0126] Distance threshold module 255 may establish a threshold for the amount or type of variation between the behavioral profile and current usage patterns. Variations in usage patterns from variation determination module 250 may be received into distance threshold module 255, which then calculates a distance of the current usage behavior from the behavioral profile. The threshold may be derived from compliance module 280, and/or from business rules, regulations, or other rules. The threshold may be based on the level of risk of requested activities in the session. For example, the distance threshold for purchasing automobile insurance may be much higher than the distance threshold for transferring funds out of an account (i.e. the variations in behavioral patterns can be much greater for purchasing insurance than for transferring funds).
[0127] Challenge level module 260 can determine a challenge level for the user and may implement the challenge level for the user to proceed in the session. The challenge level may be based on the variations between the current behavioral characteristics and historical behavioral characteristics embodied in the behavioral profile. The challenge level may be directly correlated with the distance between the current usage patterns and the behavioral profile estimated by distance threshold module 255. Accordingly, if the risk level of the activity is higher and the usage patterns are not matching up, the challenge level may be higher.
[0128] Challenge level module 260 may implement a challenge level to proceed in the session. The challenge levels may result in allowing the user to proceed with the session, collecting additional identifying information, noting suspicious activity, or disallowing the user to proceed with the session. In some embodiments, the challenge level may require a response from the user, or the challenge level may be met by information being collected from the user's device with no response or action by the user (e.g., the user is identified, verified, or authenticated passively). In other embodiments, the challenge level may be met by a next action of the user, such as the next webpage the user navigates to.
[0129] Validation module 265 validates the user and/or the behavioral profile during the current session when a usage pattern during the current session is within a predetermined distance from the typical usage pattern of the user or when the user has provided sufficient additional identifying information. Validation module 265 may receive an implemented challenge level from the challenge level module 260. The validation may be done passively, that is, without any user information, particularly when the challenge level results in allowing the user to proceed with the session. For example, the behavioral profile may validate the user's mobile device, allowing the user to take some actions without providing any additional information. Validation module 265 may also validate the user when the user provides additional information in response to a challenge level requiring the user to provide additional identifying information, if the information provided by the user satisfies the challenge level.
[0130] Response module 270 can request additional identifying information from the user. The challenge level may be received from the challenge level module 260 and the user may be presented with a question or an action that the user is required to take in order to proceed in the session. In some embodiments, the user may be providing a response without knowledge of the response. For example, the user may be required not to go to a particular website or to take another action in order to validate the behavioral profile and/or validate that the user is not a fraud. After a response to the challenge level is received, behavioral profile adaptation module 275 may be adapted based on the response.
[0131] Behavioral profile adaptation module 275 can adapt the behavioral profile of a user as additional information is gained. For example, a behavioral profile may initially be created using demographic data of users similar to the users. However, the behavioral profile may be adapted with the user behavior from the current session. The demographic information may, over time, be removed or deemphasized from the behavioral profile as the user behavioral profile is adapted with the user behavior from additional sessions with the user.
[0132] Behavioral profile adaptation module 275 may also adapt the behavioral profile based on the behavioral characteristics that triggered the challenge level. Behavioral profile adaptation module 275 may adapt the behavioral profile of the user based on responses the user provides to response module 270 when responding to a challenge level. For example, if the user is asked to provide additional identifying information, the user's behavioral profile may be updated with the information.
[0133] Compliance module 280 measures the success of the behavioral profiler in detecting fraud. In some embodiments, the success is measured by meeting baseline requirements such as computational performance (running in approximately 200 ms per page request).
[0134] In some embodiments, compliance module 280 generates a model that correctly identifies as many fraudulent transactions as possible subject to some predetermined maximum threshold on false-positives. An advantage of this approach is that it is simple and does not require a great deal of data to use.
[0135] In other embodiments, compliance module 280 uses a cost-based approach to measuring success. Varying amounts of risk are associated with different transactions. As such, the magnitude of these risks may be used to more accurately define success. For instance, with the appropriate data on the quantity of money lost to various kinds of fraudulent transactions, the model may be improved to target those transactions which pose the greatest monetary risk to the association. Likewise, if the cost of auditing or investigating transactions is known, this too can be used in decided whether to deny and flag a transaction for review. For instance, transactions with expected losses that are less than the audit and investigation costs may not be flagged for review.
[0136] Compliance module 280 may provide an optimization goal. The optimization goal may be used in the variation determination module 250 and distance threshold module 255. For example, the method of choosing numerical weights to assign to different kinds of anomalies will be different if the optimization goal is to maximize correct predictions versus maximizing the dollar value of model where the cost of fraudulent transactions is weighed against the cost of auditing false-positives.
[0137] GUI generation module 285 can generate one or more GUI screens that allow for interaction with a user. In at least one embodiment, GUI generation module 285 generates a graphical user interface allowing a user to view accounts, report losses, track claims status, set preferences, and/or otherwise receive or convey information to the user.
[0138] FIG. 5 is a flow chart illustrating a set of operations 500 for determining whether a user may engage in an activity in accordance with various embodiments of the present disclosure. The operations can be performed by various components such as processor(s) 220, channel communication module 225, activity request receiver 230, information receiving/extracting module 235, behavioral biometrics module 240, behavioral profile generation module 245, variation determination module 250, distance threshold module 255, challenge level module 260, validation module 265, response module 270, behavioral profile adaptation module 275, compliance module 280, and/or other modules or components.
[0139] Engaging operation 510 engages in a contact with a user. The contact may be through a channel such as a web portal, a mobile application, instant messaging system, face-to-face contact, and the like. The contact may include a request for the user to engage in an activity such as view an account, purchase insurance, file a claim, transfer funds, and the like. The contact with the user may establish a current session with the user. Engaging operation 510 may record, observe, and/or monitor interactions during and after the contact. Usage patterns may be detected during engaging operation 510.
[0140] Information relating to the channel, the user, and the requested activity may be collected, gathered, or otherwise extracted by collecting operation 520. Collected information may include information related to the device, location of the user, demographics of others similar to the user, prior and current web behaviors of the user (e.g., clicktrails, logon attempts), prior activity requests, security notes in the user's account, type of the requested activity, and other information.
[0141] Behavioral profile generation operation 530 uses the information collected in collecting operation 520 to generate a behavior profile for the user. Many different types of models or techniques may be used in creating the behavior profile for the user such as a Bayesian network, statistical-based anomaly detection techniques, one or more Markov models, knowledge-based techniques, neural networks, clustering and outlier detection, demographic analysis, genetic algorithms, or fuzzy logic techniques. The behavior profile is updated with information from the current session, further establishing patterns of the user.
[0142] In some embodiments, behavioral profile generation operation 530 generates the profile over time from various interactions with the user. Demographic information of users similar to the user may be incorporated into a profile for the user, particularly when little information is available.
[0143] Variations are identified between current usage patterns and the behavioral profile in identification operation 540. A threshold, or distance, between the current usage patterns and the behavioral profile may be established. A challenge level for the user to proceed with the session may be implemented based on the distance between the current usage patterns and the behavioral profile.
[0144] Decision operation 550 decides whether the user is allowed to engage in the activity and/or proceed with the session, reject the user request to engage in the activity and/or proceed in the session, or request the user to provide additional information. The various decisions in decision operation 550 are based on the variations identified in identification operation 540. For example, if variations between the behavioral profile and the current usage are minimal and thus the distance is short, the user may be permitted to continue without providing additional information. In some embodiments, if variations are great, but the risk level of the requested activity is extremely low, the user may be permitted to continue without providing additional information.
[0145] FIG. 6 is a flow chart illustrating a set of operations 600 for verifying a user in accordance with various embodiments of the present disclosure. The operations can be performed by various components such as processor(s) 220, channel communication module 225, activity request receiver 230, information receiving/extracting module 235, behavioral biometrics module 240, behavioral profile generation module 245, variation determination module 250, distance threshold module 255, challenge level module 260, validation module 265, response module 270, behavioral profile adaptation module 275, compliance module 280, and/or other modules or components.
[0146] User identification operation 610 identifies a user interacting through a channel. A channel may be a personal computer, mobile device, telephone, POS device, ATM machine, and related software and/or hardware such as a web portal, a mobile application, instant messaging system, and the like. User identification operation 610 may identify the user by a username/password, device, voice or behavioral biometrics, and the like. The interaction may include speaking with a representative regarding opening an account, opening up a new line of credit via a mobile application, transferring funds through the web portal, etc.
[0147] The identification may be passive, meaning that the user may not have to take an action to be identified. For example, the user may be identified by calling in using a phone listed in the user's profile. In another example, the user may view their account on a mobile device, in which the user is identified by the phone number associated with the device and the network accessed by the device.
[0148] After the user is identified in user identification operation 610, behavioral profile retrieving operation 620 retrieves a behavior profile associated with the user. The behavioral profile may be stored in a database such as behavioral profile database 150.
[0149] Analysis operation 630 analyzes the current user interactions and behaviors and compares these interactions with the interactions and behaviors that are predicted based on the behavioral profile. Analysis operation 630 may identify differences in current user interactions and the user's behavioral profile.
[0150] Using the information provided by analysis operation 630, estimation operation 640 estimates the distance of the user's behaviors from the prior usage patterns. For example, the behavioral profile may indicate a large variation in user behavior if the user checks an account balance 95% of the time the user transfers money to any other account and failed to check the account balance during this session but still is attempting to transfer out funds.
[0151] Challenge implementation operation 650 may implement a challenge to proceed with the session based on the distance estimated by the estimation operation 640. A large distance may indicate a high risk level, thus indicating that the actions may be fraudulent. A challenge may include a request for additional information. The amount or type of information requested may depend on the distance.
[0152] For example, if the user wants to view a webpage to related to insurance purchasing, even though the user has never viewed the insurance webpage in previous interactions, business rules may evaluate the activity as low risk and the user may not be required to provide additional authentication. In another example, supposed the user requests a transfer of money from a deposit account to an investment account at a different financial institution. The information received in connection with the channel may indicate that the IP address is from South Korea. Comparing the request to the user behavioral profile may indicate that this activity is very typical for this user because she is a member of the army stationed in South Korea and consistently transfers funds of similar values to outside investment accounts. Thus, although for many users this may be considered a high risk activity requiring some intense challenges by the challenge implementation operation, for this user, it is typical, as shown by the user's behavioral profile.
Computer System Overview
[0153] Embodiments of the present disclosure include various steps and operations, which have been described above. A variety of these steps and operations may be performed by hardware components or may be embodied in machine-executable instructions, which may be used to cause a general-purpose or special-purpose processor programmed with the instructions to perform the steps. Alternatively, the steps may be performed by a combination of hardware, software, and/or firmware. As such, FIG. 7 is an example of a computer system 700 with which embodiments of the present disclosure may be utilized. According to the present example, the computer system includes a bus 705, at least one processor 710, at least one communication port 715, a main memory 720, a removable storage media 725, a read only memory 730, and a mass storage device 735.
[0154] Processor(s) 710 can be any known processor, such as, but not limited to, an Intel® Itanium® or Itanium 2® processor(s), or AMD® Opteron® or Athlon MP® processor(s), or Motorola® lines of processors. Communication port(s) 715 can be any of an RS-232 port for use with a modem-based dialup connection, a 10/100 Ethernet port, or a Gigabit port using copper or fiber. Communication port(s) 715 may be chosen depending on a network such a Local Area Network (LAN), Wide Area Network (WAN), or any network to which the computer system 700 connects.
[0155] Main memory 720 can be Random Access Memory (RAM), or any other dynamic storage device(s) commonly known in the art. Read only memory 730 can be any static storage device(s) such as Programmable Read Only Memory (PROM) chips for storing static information such as instructions for processor 710.
[0156] Mass storage 735 can be used to store information and instructions. For example, hard disks such as the Adaptec® family of SCSI drives, an optical disc, an array of disks such as a RAID, such as the Adaptec family of RAID drives, or any other mass storage devices may be used.
[0157] Bus 705 communicatively couples processor(s) 710 with the other memory, storage, and communication blocks. Bus 705 can be a PCI/PCI-X or SCSI based system bus depending on the storage devices used.
[0158] Removable storage media 725 can be any kind of external hard-drives, floppy drives, IOMEGA® Zip Drives, Compact Disc-Read Only Memory (CD-ROM), Compact Disc-Re-Writable (CD-RW), or Digital Video Disc-Read Only Memory (DVD-ROM).
[0159] The components described above are meant to exemplify some types of possibilities. In no way should the aforementioned examples limit the disclosure, as they are only exemplary embodiments.
Terminology
[0160] Brief definitions of terms, abbreviations, and phrases used throughout this application are given below.
[0161] The terms “connected” or “coupled” and related terms are used in an operational sense and are not necessarily limited to a direct physical connection or coupling. Thus, for example, two devices may be coupled directly, or via one or more intermediary media or devices. As another example, devices may be coupled in such a way that information can be passed therebetween, while not sharing any physical connection with one another. Based on the disclosure provided herein, one of ordinary skill in the art will appreciate a variety of ways in which connection or coupling exists in accordance with the aforementioned definition.
[0162] The phrases “in some embodiments,” “according to some embodiments,” “in the embodiments shown,” “in other embodiments,” “embodiments,” and the like generally mean the particular feature, structure, or characteristic following the phrase is included in at least one embodiment of the present disclosure, and may be included in more than one embodiment of the present disclosure. In addition, such phrases do not necessarily refer to the same embodiments or different embodiments.
[0163] If the specification states a component or feature “may,” “can,” “could,” or “might” be included or have a characteristic, that particular component or feature is not required to be included or have the characteristic.
[0164] The term “responsive” includes completely or partially responsive.
[0165] The term “module” refers broadly to a software, hardware, or firmware (or any combination thereof) component. Modules are typically functional components that can generate useful data or other output using specified input(s). A module may or may not be self-contained. An application program (also called an “application”) may include one or more modules, or a module can include one or more application programs.
[0166] The term “network” generally refers to a group of interconnected devices capable of exchanging information. A network may be as few as several personal computers on a Local Area Network (LAN) or as large as the Internet, a worldwide network of computers. As used herein, “network” is intended to encompass any network capable of transmitting information from one entity to another. In some cases, a network may be comprised of multiple networks, even multiple heterogeneous networks, such as one or more border networks, voice networks, broadband networks, financial networks, service provider networks, Internet Service Provider (ISP) networks, and/or Public Switched Telephone Networks (PSTNs), interconnected via gateways operable to facilitate communications between and among the various networks.
[0167] Embodiments of the present disclosure may be provided as a computer program product which may include a machine-readable medium having stored thereon instructions which may be used to program a computer (or other electronic devices) to perform a process. The machine-readable medium may include, but is not limited to, floppy diskettes, optical discs, compact disc read-only memories (CD-ROMs), and magneto-optical discs, ROMs, random access memories (RAMs), erasable programmable read-only memories (EPROMs), electrically erasable programmable read-only memories (EEPROMs), magnetic or optical cards, flash memory, or other type of media/machine-readable medium suitable for storing electronic instructions. Moreover, embodiments of the present disclosure may also be downloaded as a computer program product, wherein the program may be transferred from a remote computer to a requesting computer by way of data signals embodied in a carrier wave or other propagation medium via a communication link (e.g., a modem or network connection).
[0168] Also, for the sake of illustration, various embodiments of the present disclosure have herein been described in the context of computer programs, physical components, and logical interactions within modern computer networks. Importantly, while these embodiments describe various embodiments of the present disclosure in relation to modern computer networks and programs, the method and apparatus described herein are equally applicable to other systems, devices, and networks as one skilled in the art will appreciate. As such, the illustrated applications of the embodiments of the present disclosure are not meant to be limiting, but instead examples. Other systems, devices, and networks to which embodiments of the present disclosure are applicable include, but are not limited to, other types of communication and computer devices and systems. More specifically, embodiments are applicable to communication systems, services, and devices such as cell phone networks and compatible devices. In addition, embodiments are applicable to all levels of computing from the personal computer to large network mainframes and servers.
[0169] In conclusion, the present disclosure provides novel systems, methods, and arrangements for behavioral profiling. While detailed descriptions of one or more embodiments of the disclosure have been given above, various alternatives, modifications, and equivalents will be apparent to those skilled in the art without varying from the spirit of the disclosure. For example, while the embodiments described above refer to particular features, the scope of this disclosure also includes embodiments having different combinations of features and embodiments that do not include all of the described features. Accordingly, the scope of the present disclosure is intended to embrace all such alternatives, modifications, and variations as fall within the scope of the claims, together with all equivalents thereof. Therefore, the above description should not be taken as limiting.
Claims
What is claimed is:
1. A method of user verification comprising:
observing, by a computer processor, behavioral characteristics of user interactions during a current session with a user through one of a plurality of channels;
identifying, in real-time or near real-time, variations between the behavioral characteristics of the user interactions observed during the current session and a behavioral profile previously developed based on prior usage patterns of the user through the plurality of channels, wherein the behavioral profile is based on clicktrail data and authentication logs;
implementing, by the computer processor, a challenge level to proceed in the current session, the challenge level based on the variations between the behavioral characteristics and the behavioral profile and on a risk level of requested activities of the current session; and
when the challenge level to proceed in the current session disallows the user to continue in the current session:
analyzing, by the computer processor, behavioral biometrics of the user,
comparing, by the computer processor, the behavioral biometrics of the user with one or more previous samples of the behavioral biometrics of the user, and
when the comparison of the behavioral biometrics of the user with the one or more previous samples of the behavioral biometrics of the user is within a tolerance, allowing, by the computer processor, the current session to proceed without requesting a user response or notifying the user of the challenge level.
2. The method of claim 1, further comprising receiving current device information, and wherein identifying the variations between the behavioral characteristics and the behavioral profile includes comparing the current device information with historical device information stored in the behavioral profile.
3. The method of claim 2, wherein the current device information includes at least one of the following: device location, device identification, channel usage on a current device, language, network, or internet service provider.
4. The method of claim 1, wherein identifying the variations includes estimating a distance between the behavioral characteristics in the current session and the behavioral profile, wherein the behavioral profile was further developed based on website behaviors specific to a generation of the user.
5. The method of claim 1, the method further comprising:
developing the behavioral profile by identifying typical usage patterns of behavior of the user from historical usage data;
calculating a distance between the behavioral characteristics in the current session and the behavioral profile; and
validating the behavioral profile during the current session when the behavioral characteristics in the current session are within a predetermined distance from the typical usage patterns of behavior of the user.
6. The method of claim 1, wherein the behavioral profile includes the behavioral characteristics, and wherein the method further comprises representing at least a portion of the behavioral characteristics as metrics.
7. The method of claim 1, wherein the behavioral profile includes the behavioral characteristics, and wherein the behavioral characteristics include at least one of: a duration between keyboard strokes, an intensity of keyboard presses, a user central processing unit time, a system central processing unit time, an amount of time between user logons, an amount of character IO during application execution, a maximum size of a request, a real user usage over time, a virtual size of text segment, a total number of times opened during the application execution, a number of major and minor page faults, a number of pages accessed, a measure of disk IO, an elapsed time in seconds of the current session, a number of signals received during the application execution, a name of a new user name changed during the current session, a number of unique internet protocol addresses used per month, an hour of day of the current session, a number of distinct pages accessed during the current session, whether an application executed was executed on a remote network host, a number of computers used each month, a name of the remote network host if the application was executed on the remote network host, a name of a local network host on which the application was executed, a number of different users with a same internet protocol address, a number of seconds since a last audit record for the application, or a number of times the user logs in per a predetermined time period.
8. The method of claim 1, the method further comprising:
adapting the behavioral profile based on the behavioral characteristics of the user interactions observed during the current session.
9. The method of claim 8, wherein the behavioral profile is initially created using demographic data of users similar to the user.
10. The method of claim 9, the method further comprising:
removing or deemphasizing at least a portion of the demographic data from the behavioral profile as the behavioral profile of the user is adapted with the behavioral characteristics of the user interactions observed during the current session.
11. The method of claim 1, the method further comprising:
receiving a response to the challenge level;
determining whether the response to the challenge level authenticates the user; and
adapting the behavioral profile based on the behavioral characteristics that triggered the challenge level.
12. The method of claim 11, wherein the challenge level includes at least one of:
allowing the user to proceed with the current session, collecting identifying information, noting suspicious activity, or disallowing the user to proceed with the current session.
13. The method of claim 1, wherein the variations are indicative of a second user, and wherein the method further comprises:
determining that the second user is authorized by the user; and
developing a behavioral profile for the second user.
14. The method of claim 1, wherein the behavioral profile is further based on previous warnings indicating suspicious activity.
15. The method of claim 1, wherein the plurality of channels comprises an internet portal, face-to-face contact, a mobile application, and an instant messaging system.
16. The method of claim 1, wherein the behavioral profile is developed using at least one of the following: Bayesian network, statistical-based anomaly detection techniques, one or more Markov models, knowledge-based techniques, neural networks, clustering and outlier detection, demographic analysis, genetic algorithms, or fuzzy logic techniques.
17. A computer-implemented method of fraud prediction comprising:
passively identifying a user interacting through a channel during a current session;
retrieving, from a database, a predictive behavioral profile associated with the user, wherein the predictive behavioral profile receives current user interactions with the channel and estimates a distance from prior usage patterns of the user, and wherein the predictive behavioral profile is based on clicktrail data and authentication logs;
identifying, by a computer processor, in real-time or near real-time, variations between current usage patterns of the user and the predictive behavioral profile;
calculating similarity or distance measures between the current session of the user and the predictive behavioral profile;
translating a set of the calculated similarity or distance measures into a single confidence measure;
implementing a challenge level to proceed in the current session, wherein the challenge level is based on a risk level of requested activities of the current session; and
when the challenge level to proceed in the current session disallows the user to continue in the current session:
analyzing, by the computer processor, behavioral biometrics of the user,
comparing, by the computer processor, the behavioral biometrics of the user with one or more previous samples of the behavioral biometrics of the user, and
when the comparison of the behavioral biometrics of the user with the one or more previous samples of the behavioral biometrics of the user is within a tolerance, allowing, by the computer processor, the current session to proceed without requesting a user response or notifying the user of the challenge level.
18. The computer-implemented method of claim 17, further comprising developing the predictive behavioral profile using at least one of the following: Bayesian network, statistical-based anomaly detection techniques, one or more Markov models, knowledge-based techniques, neural networks, clustering and outlier detection, demographic analysis, genetic algorithms, or fuzzy logic techniques.
19. A system for authenticating a user, the system comprising:
a memory; and
a processor in communication with the memory, the processor configured to execute software modules, the software modules comprising:
a channel communication module configured to engage in one or more sessions with a user via a plurality of channels;
an information gathering module configured to:
monitor user behavior during the one or more sessions; and
collect demographic data relating to the user;
a behavioral profile generation module configured to:
develop a user behavioral profile based on the user behavior, the user behavioral profile including patterns of behavior that are typical of the user, wherein the user behavioral profile is based on clicktrail data and authentication logs;
the channel communication module further configured to observe the user behavior during a session;
a variation determining module configured to determine, in near real-time, variations between the user behavior observed during the session and the user behavioral profile; and
a challenge module configured to:
implement a challenge level to proceed with the session based on the variations, wherein the challenge level is based on a risk level of requested activities of the session;
adapt the user behavioral profile with the user behavior from the session; and
when the challenge level to proceed in the session disallows the user to continue in the session:
analyzing behavioral biometrics of the user,
comparing the behavioral biometrics of the user with one or more previous samples of the behavioral biometrics of the user, and
when the comparison of the behavioral biometrics of the user with the one or more previous samples of the behavioral biometrics of the user is within a tolerance, allowing the session to proceed without requesting a user response or notifying the user of the challenge level.
20. The system of claim 19, wherein the user behavioral profile is further based on website behaviors specific to a generation of the user.
Patent Citations (15)
| Patent | Date | Inventor | Cited By |
|---|---|---|---|
| US6252869(B1) | 2001-06-01 | Silverman | Applicant |
| US7389917(B2) | 2008-06-01 | Abraham et al. | Applicant |
| US7490242(B2) | 2009-02-01 | Torres et al. | Applicant |
| US7908645(B2) | 2011-03-01 | Varghese et al. | Examiner |
| US2004/0006532(A1) | 2004-01-01 | Lawrence et al. | Applicant |
| US2009/0018904(A1) | 2009-01-01 | Shipman et al. | Examiner |
| US2009/0199264(A1) | 2009-08-01 | Lang | Applicant |
| US2009/0254971(A1) | 2009-10-01 | Herz et al. | Examiner |
| US2011/0016534(A1) | 2011-01-01 | Jakobsson et al. | Applicant |
| US2012/0030047(A1) | 2012-02-01 | Fuentes et al. | Examiner |
| US2012/0054057(A1) | 2012-03-01 | O'Connell et al. | Examiner |
| US2012/0054646(A1) | 2012-03-01 | Hoomani et al. | Examiner |
| US2013/0123007(A1) | 2013-05-01 | Williams et al. | Applicant |
| US2013/0239191(A1) | 2013-09-01 | Bostick et al. | Applicant |
| WO-2008061066 | 2008-05-01 | Applicant |
Non-Patent Literature (78)
- Albrecht, Conan C. et al., “Conducting a Pro-Active Fraud Audit: A Case Study”, Journal of Forensic Accounting, 2000, pp. 203-218, vol. II, R.T. Edwards, Inc.Applicant
- Albrecht, Conan C. et al., “Strategic Fraud Detection: A Technology-Based Model”, 2010, retrieved from http://www.business.umt.edu/Libraries/Seminar—Series/F04-Strategic—Fraud—Detection.sflb.ashx.Applicant
- Anderson, Debra et al., “Detecting Unusual Program Behavior Using the Statistical Component of the Next-generation Intrusion Detection Expert System (NIDES)”, SRI International Technical Report SRI-CSL-95-06, May 1995, 86 pages, Computer Science Laboratory.Applicant
- Anderson, Debra et al., “Next Generation Intrusion Detection Expert System (NIDES), Software Users Manual, Beta-Update Release”, SRI International Technical Report SRI-CSL-95-0, Dec. 1, 1994, 316 pages, Computer Science Laboratory, Menlo Park, CA.Applicant
- Barson, P. et al., “The Detection of Fraud in Mobile Phone Networks”, Neural Network World, 1996, pp. 477-484, vol. 6, No. 4.Applicant
- Bell, Timothy B. et al., “A Decision Aid for Assessing the Likelihood of Fraudulent Financial Reporting”, Auditing: A Journal of Practice & Theory, Spring 2000, pp. 68-84, vol. 19, No. 1.Applicant
- Berry, Michael et al., “Data Mining Techniques: Theory and Practice”, 2009, ISBN 978-1-60764-366-1, SAS Institute.Applicant
- Boatsman, James R. et al., “The Effects of Decision Consequences on Auditors' Reliance on Decision Aids in Audit Planning”, Organizational Behavior and Human Decision Processes, Aug. 1997, pp. 211-247, vol. 71, No. 2, Academic Press.Applicant
- Bolton, Richard J. et al., “Statistical Fraud Detection: A Review”, Statistical Science, 2002, pp. 235-255, vol. 17, No. 3.Applicant
- Bonifacio, Jr., J. M. et al., “An Adaptive Intrusion Detection System Using Neural Networks”, Proceedings of the 14th International Information Security Conference (IFIP/Sec'98, part of the 15th IFIP World Computer Congress), Aug. 31-Sep. 4, 1998, pp. 276-280, ISBN: 3-85403-116-5, IFIP, Austrian Computer Society, Vienna, Austria and Budapest, Hungary.Applicant
- Brause, R. et al., “Neural Data Mining for Credit Card Fraud Detection”, In Tools with Artificial Intelligence, Proceedings of the 11th IEEE International Conference, 1999, pp. 103-106, IEEE.Applicant
- Breunig, Markus M. et al., “LOF: Identifying Density-Based Local Outliers”, Proceedings of the ACM SIGMOD International Conference on Management of Data, 2000, pp. 93-104, Dallas, TX.Applicant
- Bridges, Susan M. et al., “Fuzzy Data Mining and Genetic Algorithms Applied to Intrusion Detection”, Proceedings of the National Information Systems Security Conference (NISSC), Oct. 16-19, 2000, pp. 13-31, Baltimore, MD.Applicant
- Burge, Peter et al., “An Unsupervised Neural Network Approach to Profiling the Behavior of Mobile Phone Users for Use in Fraud Detection”, Journal of Parallel and Distributed Computing, May 10, 2001, pp. 915-925, vol. 61, Academic Press.Applicant
- Casella, George et al., “Statistical Inference—Second Edition”, 2001, 2 pages, ISBN: 0-534-24312-6, Duxbury, Pacific Grove, CA.Applicant
- Chou, Youn-Min et al., “Power Comparisons for a Hotelling's T2 Statistic”, Communications in Statistics—Simulation and Computation, 1999, pp. 1031-1050, vol. 28, No. 4.Applicant
- Cortes, Corinna et al., “Signature-Based Methods for Data Streams”, Data Mining and Knowledge Discovery, 2001, pp. 167-182, vol. 5, Kluwer Academic Publishers, The Netherlands.Applicant
- Cox, Kenneth C. et al., “Visual Data Mining: Recognizing Telephone Calling Fraud”, Data Mining and Knowledge Discovery, Apr. 17, 1997, pp. 225-231.Applicant
- Dasgupta, Dipankar et al., “An Immunity-Based Technique to Characterize Intrusions in Computer Networks”, IEEE Transactions on Evolutionary Computation, Jun. 2002, pp. 281-291, vol. 6, No. 3, IEEE.Applicant
- Debar, Herve et al., “A Neural Network Component for an Intrusion Detection System”, IEEE Symposium on Research in Computer Security and Privacy, 1992, pp. 240-250, IEEE.Applicant
- Decker, Paul, “Data's Mining Hidden Dangers”, Banking Strategies, Mar.-Apr. 1998, pp. 6-14.Applicant
- Denning, Dorothy E., “An Intrusion-Detection Model”, IEEE Transactions on Software Engineering, Feb. 1987, pp. 222-232, vol. SE-13, No. 2.Applicant
- Dickerson, John E. et al., “Fuzzy Network Profiling for Intrusion Detection”, Proceedings of the 19th International Conference of the North American Fuzzy Information Processing Society (NAFIPS), 2000, pp. 301-306.Applicant
- Dokas, Paul et al., “Data Mining for Network Intrusion Detection”, Proceedings of the NSF Workshop on Next Generation Data Mining, Nov. 2002, pp. 21-30.Applicant
- Estevez, Pablo A. et al., “Subscription fraud prevention in telecommunications using fuzzy rules and neural networks”, Expert Systems with Applications, 2006, pp. 337-344, vol. 31, Elsevier Ltd.Applicant
- Estevez-Tapiador, Juan M. et al., “Anomaly detection methods in wired networks: a survey and taxonomy”, Computer Communications, 2004, pp. 1569-1584, vol. 27, Elsevier B.V.Applicant
- Estevez-Tapiador, Juan M. et al., “Detection of Web-based Attacks through Markovian Protocol Parsing”, Proceedings of the 10th IEEE Symposium on Computers and Communications (ISCC 2005), 2005, pp. 457-462, IEEE.Applicant
- Estevez-Tapiador, Juan M. et al., “Stochastic Protocol Modeling for Anomaly-Based Network Intrusion Detection”, Proceedings of the 2003 IEEE International Workshop on Information Assurance, Mar. 24, 2003, pp. 3-12, ISBN 0-7695-1886-9, IEEE Press.Applicant
- Everitt, Brian S., “A Monte Carlo Investigation of the Robustness of Hotelling's One- and Two-Sample T2 Tests”, Journal of the American Statistical Association, Mar. 1979, pp. 48-51, vol. 74, No. 365, American Statistical Association.Applicant
- Ezawa, Kazuo J. et al., “Learning Goal Oriented Bayesian Networks for Telecommunications Risk Management”, Proceedings of the 13th International Conference on Machine Learning, 1996, pp. 139-147, Morgan Kaufmann.Applicant
- Fox, Kevin L. et al., “A Neural Network Approach Towards Intrusion Detection”, 13th National Computer Security Conference, Oct. 1-4, 1990, pp. 125-134, Washington, D.C.Applicant
- Frank, R.J. et al., “Time Series Prediction and Neural Networks”, Journal of Intelligent & Robotic Systems, 2001, pp. 91-103, vol. 31, No. 1.Applicant
- Garcia-Teodoro, P. et al., “Anomaly-based network intrusion detection: Techniques, systems and challenges”, Computers & Security, 2009, pp. 18-28, vol. 28, Elsevier Ltd.Applicant
- Green, Brian Patrick et al., “Assessing the Risk of Management Fraud Through Neural Network Technology”, Auditing: A Journal of Practice & Theory, Spring 1997, pp. 14-28, vol. 16, No. 1.Applicant
- Heckerman, David, “A Tutorial on Learning With Bayesian Networks”, Technical Report MSR-TR-95-06, Nov. 1996, 58 pages, Microsoft Research, Redmond, WA.Applicant
- Huang, Ling et al., “Adversarial Machine Learning”, Proceedings of the 4th ACM Workshop on Security and Artificial Intelligence, Oct. 21, 2011, pp. 43-58, ACM, Chicago, IL.Applicant
- Jiang, Cheng-Huang et al., “Keystroke Statistical Learning Model for Web Authentication”, ASIACCS '07 Proceedings of the 2nd ACM Symposium on Information, Computer and Communications Security, Mar. 20-22, 2007, pp. 359-361, ACM, Singapore.Applicant
- Khattree, Ravindra et al., “Multivariate Data Reduction and Discrimination with SAS© Software”, 2000, SAS Institute.Applicant
- Knapp, Carol A. et al., “The effects of experience and explicit fraud risk assessment in detecting fraud with analytical procedures”, Accounting, Organizations and Society, 2001, pp. 25-37, vol. 26, Elsevier Science Ltd.Applicant
- Kruegel, Christopher et al., “Bayesian Event Classification for Intrusion Detection”, Proceedings of the 19th Annual Computer Security Applications Conference, 2003.Applicant
- Lazarevic, Aleksandar et al., “Intrusion Detection: A Survey”, Managing Cyber Threats: Issues, Approaches, and Challenges, 2005, 60 pages, Springer Verlag.Applicant
- Lee, Wenke et al., “Real Time Data Mining-based Intrusion Detection”, DARPA Information Survivability Conference & Exposition II, Proceedings of DISCEX'01, 2001, pp. 89-100, vol. 1, IEEE.Applicant
- Li, Wei, “Using Genetic Algorithm for Network Intrusion Detection”, 2004, pp. 1-8, C.S.G. Department of Energy.Applicant
- Liao, Yihua et al., “Using K-Nearest Neighbor Classifier for Intrusion Detection”, Computers & Security, 2002, pp. 439-448, vol. 21.Applicant
- Lin, Jerry W. et al., “A fuzzy neural network for assessing the risk of fraudulent financial reporting”, Managerial Auditing Journal, 2003, pp. 657-665, vol. 18, No. 8, MCB UP Limited.Applicant
- Litan, Avivah, “Magic Quadrant for Web Fraud Detection”, Apr. 19, 2011, 37 pages, Gartner, Inc., ID No. G00212116.Applicant
- Lunt, Teresa F. et al., “A Real-Time Intrusion-Detection Expert System (IDES)”, SRI International Final Technical Report, Feb. 28, 1992, 166 pages, Menlo Park, CA.Applicant
- Mackay, David J.C., “Information Theory, Inference, and Learning Algorithms”, 2003, 640 pages, ISBN 0-521-64298-1, Cambridge University Press.Applicant
- Maes, Sam et al., “Credit Card Fraud Detection Using Bayesian and Neural Networks”, American Association of Neurological Surgeons, 1993, pp. 261-270.Applicant
- Mahoney, Matthew V. et al., “An Analysis of the 1999 DARPA/Lincoln Laboratory Evaluation Data for Network Anomaly Detection”, Technical Report CS-2003-02, 2003, 20 pages, Florida Institute of Technology, Melbourne, FL.Applicant
- Major, John A. et al., “EFD: A Hybrid Knowledge/Statistical-Based System for the Detection of Fraud”, International Journal of Intelligent Systems, 1992, pp. 687-703, vol. 26, John Wiley & Sons, Inc.Applicant
- Mason, R. L. et al., “Decomposition of T2 for Multivariate Control Chart Interpretation”, Journal of Quality Technology, 1995, pp. 99-108, vol. 27, No. 2.Applicant
- Mason, Robert L. et al., “A Practical Approach for Interpreting Multivariate T2 Control Chart Signals”, Journal of Quality Technology, Oct. 1997, pp. 396-406, vol. 29, No. 4.Applicant
- Mason, Robert L. et al., “Assessment of Multivariate Process Control Techniques”, Journal of Quality Technology, Apr. 1997, pp. 140-143, vol. 29, No. 2.Applicant
- Mason, Robert L. et al., “Improving the Sensitivity of the T2 Statistic in Multivariate Process Control”, Journal of Quality Technology, Apr. 1999, pp. 155-164, vol. 31, No. 2.Applicant
- Murad, Uzi et al., “Unsupervised Profiling for Identifying Superimposed Fraud”, Principles of Data Mining and Knowledge Discovery, Proceedings of Third European Conference, PKDD'99, 1999, pp. 264-274, Springer-Verlag, Berlin Heidelberg.Applicant
- Nieschwietz, Robert J. et al., “Empirical Research on External Auditors' Detection of Financial Statement Fraud”, Journal of Accounting Literature, 2000, pp. 190-246, vol. 19.Applicant
- Nigrini, Mark J., “Forensic Analytics: Methods and Techniques for Forensic Accounting Investigations”, 2011, ISBN 978-0-470-89046-2, John Wiley & Sons, Inc., Hoboken, NJ.Applicant
- Pak, Alexander et al., “Twitter as a Corpus for Sentiment Analysis and Opinion Mining”, Proceedings of LREC, May 2010, 7 pages.Applicant
- Palshikar, Girish Keshav, “The Hidden Truth”, Intelligent Enterprise, May 28, 2002, pp. 46-51, vol. 5, No. 9.Applicant
- Patcha, Animesh et al., “An Overview of Anomaly Detection Techniques: Existing Solutions and Latest Technological Trends”, Computer Networks, 2007, pp. 3448-3470, vol. 51, No. 12, DOI: 10.1016/j.comnet.2007.02.001.Applicant
- Phua, Clifton et al., “A Comprehensive Survey of Data Mining-based Fraud Detection Research”, 2005, Clayton School of Information Technology, Monash University, Clayton, Victoria, Australia.Applicant
- Porras, Phillip A. et al., “Emerald: Event Monitoring Enabling Responses to Anomalous Live Disturbances”, 20th NIST-NCSC National Information Systems Security Conference, 1997, pp. 353-365, Baltimore, MD.Applicant
- Portnoy, Leonid et al., “Intrusion Detection with Unlabeled Data Using Clustering”, Proceedings of the ACM Workshop on Data Mining Applied to Security, 2001, 14 pages.Applicant
- Ramadas, Manikantan et al., “Detecting Anomalous Network Traffic with Self-organizing Maps”, Recent Advances in Intrusion Detection, Proceedings of 6th International Symposium, RAID 2003, 2003, pp. 45-63, Springer-Verlag, Berlin Heidelberg.Applicant
- Sekar, R. et al., “Specification-based Anomaly Detection: A New Approach for Detecting Network Intrusions”, Proceedings of the Ninth ACM Conference on Computer and Communications Security (CCS '02), Nov. 18-22, 2002, pp. 265-274, ACM, Washington, D.C.Applicant
- Sequeira, Karlton David, “Anomaly-Based Data Mining of Intrusions”, Proceedings of the 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2002, pp. 386-395.Applicant
- Shelton, Sandra Waller et al., “Auditing Firms' Fraud Risk Assessment Practices”, Accounting Horizons, Mar. 2001, pp. 19-33, vol. 15, No. 1, American Accouting Association.Applicant
- Smaha, Stephen E., “Haystack: An Intrusion Detection System”, Aerospace Computer Security Applications Fourth Conference, 1988, pp. 37-44, IEEE.Applicant
- Stolfo, Salvatore J. et al., “Cost-based Modeling for Fraud and Intrusion Detection: Results from the JAM Project”, DARPA Information Survivability Conference and Exposition, Proceedings of DISCEX'00, 2000, pp. 130-144, vol. 2, IEEE.Applicant
- Tombini, Elvis et al., “A Serial Combination of Anomaly and Misuse IDSes Applied to HTTP Traffic”, 20th Annual Computer Security Applications Conference, 2004, Tucson, AZ.Applicant
- Viaene, Stijn et al., “A Comparison of State-of-the-Art Classification Techniques for Expert Automobile Insurance Claim Fraud Detection”, The Journal of Risk and Insurance, 2002, pp. 373-421, vol. 69, No. 3.Applicant
- Vigna, Giovanni et al., “The STAT Tool Suite”, DARPA Information Survivability Conference and Exposition, Proceedings of DISCEX'00, 2000, pp. 46-55, vol. 2, IEEE.Applicant
- Ye, Nong et al., “Multivariate Statistical Analysis of Audit Trails for Host-Based Intrusion Detection”, IEEE Transactions on Computers, Jul. 2002, pp. 810-820, vol. 51, No. 7.Applicant
- Ye, Nong et al., “Robustness of the Markov-Chain Model for Cyber-Attack Detection”, IEEE Transactions on Reliability, Mar. 2004, pp. 116-123, vol. 53, No. 1, IEEE.Applicant
- Yeung, Dit-Yan et al., “Host-Based Intrusion Detection Using Dynamic and Static Behavioral Models”, Pattern Recognition, 2003, pp. 229-243, vol. 36, No. 1.Applicant
- Zhang, Jiong et al., “A Hybrid Network Intrusion Detection Technique Using Random Forests”, Proceedings of the First International Conference on Availability, Reliability and Security (ARES'06), 2006, pp. 262-269, IEEE.Applicant
- Barbacci, Mario et al., “Quality Attributes”, CMU/SEI Technical Report, Dec. 1995, referenced online at http://www.sei.cmu.edu/pub/documents/95.reports/pdf/tr021.95.pdf.Applicant