SYSTEM AND METHOD FOR REAL-TIME TRANSACTIONAL DATA OBFUSCATION

[0001] This application claims the benefit of priority to U.S. Provisional Patent Application No. 61/369,000, titled “SYSTEM AND METHOD FOR REAL-TIME TRANSACTIONAL DATA OBFUSCATION”, filed Jul. 29, 2010; which application is herein incorporated by reference.

[0002] A portion of the disclosure of this patent document contains material which is subject to copyright protection. The copyright owner has no objection to the facsimile reproduction by anyone of the patent document or the patent disclosure, as it appears in the Patent and Trademark Office patent file or records, but otherwise reserves all copyright rights whatsoever.

[0003] The present invention is generally related to computer transactions, and is particularly related to a system and method for providing transactional data privacy while maintaining data usability, including the use of different obfuscation functions for different data types to securely obfuscate the data, in real-time, while maintaining its statistical characteristics.

[0004] New data privacy laws have appeared recently, such as the HIPAA laws for protecting medical records, and the PCI guidelines for protecting credit card information. Data privacy can be defined as maintaining the privacy of Personal Identifiable Information (PII) from unauthorized accessing. PII includes any piece of data that can be used alone, or in conjunction with additional information, to uniquely identify an individual. Examples of such information include national identification numbers, credit card numbers, as well as financial and medical records. Access control methods and data encryption provide a level of data protection from unauthorized access. However, it is not enough—for example, it does not prohibit identity thefts. It was recently suggested that 70% of data privacy breaches are internal breaches that involve an employee from the enterprise who has access to some training or testing database replica, which contains all the PII. Accordingly, in addition to access control, what are needed are techniques to protect such datasets, including preserving the data usability while protecting its privacy. These challenges are further complicated when realtime requirements are added. This is the general area that embodiments of the invention are intended to address.

[0005] Described herein is a system and method for providing transactional data privacy while maintaining data usability, including the use of different obfuscation functions for different data types to securely obfuscate the data, in real-time, while maintaining its statistical characteristics. In accordance with an embodiment, the system comprises an obfuscation process that captures data while it is being received in the form of data changes at a first or source system, selects one or more obfuscation techniques to be used with the data according to the type of data captured, and obfuscates the data, using the selected one or more obfuscation techniques, to create an obfuscated data, for use in generating a trail file containing the obfuscated data, or applying the data changes to a target or second system.

[0014] Data privacy is no more an optional feature—it is a requirement by any data management system to preserve the privacy of the data of the users of the system. Recently, new privacy laws have appeared, such as the HIPAA laws for protecting medical records, and the PCI guidelines for protecting credit card information. Data privacy (also referred to as information privacy) can be defined as maintaining the privacy of personal identifiable information or data from unauthorized accessing. Data privacy refers to developing relationship and interaction between technology and the privacy of personally identifiable information (PII) that is collected, stored, and shared by organizations. PII includes any piece of data that can be used alone, or in conjunction with additional information, to uniquely identify an individual. Examples of such information may include first and last names, social security numbers, national identification numbers, addresses, date of birth, phone numbers, email addresses, driver's license numbers, credit card numbers, financial, and medical records.

[0015] Data security has generally been provided through access control. Although access control methods provide a level of data protection, it is not enough. Access control methods, in addition to data encryption, protect data from unauthorized access. However, it does not prohibit identity thefts. It was recently suggested that that 70% of data privacy breaches are internal breaches that involve an employee from the enterprise who has access to some training or testing database replica, which contains all the PII. Therefore, there is a need for techniques that would prevent such identity thefts. Ideally, what is needed is a technique that would protect the PII from unauthorized access, and allow access for analysis, testing and training purposes, while maintaining its usability. The challenge here is the contradicting requirement of a usable copy of the data that, yet, does not breach the privacy of the data. Examples of systems that can benefit from these requirements may include large financial credit card enterprises.

[0016] As referred to herein, Data Obfuscation (DO) is a broad term that refers to any data manipulation technique used to induce ambiguity to the data, desensitize it to be of no sense, yet usable, and thus preserving its privacy. The main requirements of a DO technique are data privacy and usability. Data privacy refers to the fact that the PII are secured and concealed upon applying the DO technique to the data. Usability refers to the fact that the transformed data is still useful and maintains the main statistical and semantic properties of the original data. In addition, there are a set of desired properties, such as:

• • [0017] 1. Providing access to the confidential attributes should provide the intruder with no additional information. In other words, the ability to predict the original data given access to the obfuscated data should never be possible to use it to retrieve the original sensitive data given the technique and the obfuscated data.
  • [0018] 2. The DO technique should be irreversible. It should never be possible to use it to retrieve the original sensitive data given the technique and the obfuscated data.
  • [0019] 3. Semantics and referential integrity must be maintained.
  • [0020] 4. Obfuscation must be a repeatable process to guarantee consistency. This means that every time a data item is being obfuscated, it is obfuscated to the same obfuscated data item.

[0021] All of these requirements make the task of obfuscating the data efficiently a substantial challenge, which is even more challenging when realtime requirements are added as in the motivating example below.

[0022] Consider the case when a software-based data replication product, such as Oracle GoldenGate, is used to replicate bank transactional data across heterogeneous sites, where one copy of the data is replicated to a third party site to be used for real-time analysis purposes, say for fraud detection for instance. One way to do so is to replicate the data, and then apply an existing obfuscation technique in an offline fashion and then use the obfuscated copy for analysis. Note that a mapping between original and obfuscated data items is needed in this example. This can be maintained securely encrypted at the original data host. This solution, although relatively simple, does not satisfy the real-time requirements of the fraud detection. In addition, a copy of the original data is being copied and stored at a third party site before it is being obfuscated, which is a huge security threat. Thus, a need for a real-time transactional data obfuscation technique is needed: a technique that satisfies all desired properties of obfuscation techniques, in addition to satisfying the real-time requirements.

[0023] As disclosed herein, in accordance with an embodiment a system and method is disclosed for providing a transactional data obfuscation solution, which can be used, e.g. with software-based data replication products such as Oracle GoldenGate. In accordance with an embodiment, the system utilizes different obfuscation functions for different data types to securely obfuscate, on real-time, the data while maintaining statistical characteristics of the data, for testing and analysis purposes.

[0024] Many techniques have been proposed for data privacy such as: (1) data randomization, which adds noise to the data; (2) data anonymization, which uses generalization and suppression to make the data ambiguous; (3) data swapping, which involves ranking data items and swapping records that are close to each other; (4) geometric transformation, which uses transformations such as rotation, scaling, and translation for distorting the data; and (5) nearest neighbor data substitution, which uses Euclidean distance to define neighbors, and then perform swapping. Some of these techniques apply to only certain data types; for example, geometric transformation techniques apply only to numerical data. The majority of these techniques were developed for privacy protection for data mining and analysis, for which there are no real-time requirements. To the inventors knowledge, all these techniques involve an offline analysis phase, at which the statistical characteristics of the data set is captured, and used to guide the obfuscation, in order to maintain these statistical characteristics.

[0025] In some environments, such as with a software-based data replication product such as Oracle GoldenGate, transactional data is being replicated on real-time fashion, and hence, a real-time obfuscation technique is needed. In accordance with an embodiment, the system provides a suite of techniques for obfuscating different data types. For example, for numerical data, a technique is proposed herein that is based on both geometric transformation, namely GT-NeNDS, and anonymization. These two techniques are described in more detail below.

[0026] GT-NeNDS is a numerical data obfuscation technique that is designed for clustering mining. In accordance with an embodiment, GT-NeNDS is extended to make it applicable in real-time, by applying anonymization, which adds to the data privacy, and increases reversibility, at the expense of data loss. However, this data loss can be controlled, as explained in further detail below.

[0027] Anonymization techniques map multiple data items into one; for example, it can replace the date with the month and year only. This generalization involves a loss of information, but data stays consistent. K-anonymity aims at mapping at maximum k data items into one representing data item. Anonymization techniques are irreversible, since there no way to know the original data item.

[0028] GT-NeNDS stands for Geometric Transformation—Nearest Neighbor Data Substitution. GT techniques include scaling, rotating, and translation, these preserve data characteristics. The NeNDS technique was proposed for privacy preservation for Clustering Mining applications. It proceeds like this: it clusters the original dataset into sets of neighbors. Neighborhood is determined using Euclidean Distance. Each data item in a neighbors' set is replaced by the nearest neighbor in this set, in a way such that no swapping occurs, using special data structures. Thus, statistical properties of the original data are preserved. NeNDS introduce a degree of obfuscation by replacing a data item with its nearest neighbor. GT-NeNDS aims at securing the data by further obfuscating the nearest neighbor, using the GT techniques.

[0029] GT-NeNDS does not adequately fit real-time requirements due to the following reasons. First, to construct the sets of neighbors, the algorithm needs a pass through all the data, which is not feasible in real-time settings. Second, substituting a data item with its nearest neighbor means that the substitution is not repeatable because neighbors changes with insertions and deletions.

[0030] To overcome these shortages, in accordance with an embodiment, a GT-ANeNDS technique, and an extension to GT-NeNDS are disclosed herein.

[0031] FIG. 1 illustrates an environment or system which can use data obfuscation, in accordance with an embodiment. As shown in FIG. 1, the system 102 can be implemented with or as part of a software-based data replication product, such as Oracle GoldenGate (and which, in accordance with an embodiment, is referred to herein as BronzeGate). The system includes a userExit process 103, which performs user defined customized transformations to the replicated transactions. The system (i.e. BronzeGate) is hence a special type of userExit process, where the task is to perform the required obfuscation on the fly.

[0032] As further shown in FIG. 1, the process can be configured to run at an original database site 116, to obfuscate the transactional data before they are shipped to a replicate site, for example as part of a trail information 120, such as a GoldenGate trail file. The process is activated by the capture process 118, which monitors the original database. Whenever a transaction is committed to the original database, the capture process will capture this change and signals the userExit (BronzeGate) process to handle this transaction. The system then uses the parameters file 108, histograms 122, 124, and dictionaries 110, 112 to obfuscate the new transaction. Once done, the system sends the obfuscated transaction back to the capture process which simply writes it to the trail, which shall be shipped to the replication site.

[0033] In accordance with an embodiment a GT-ANeNDS technique is proposed herein, which overcomes GT-NeNDS' real-time limitations, and leverage the level of data privacy.

[0034] FIG. 2 illustrates the use of a GT-ANeNDS algorithm 130, in accordance with an embodiment. GT-ANeNDS combines anonymization and NeNDS techniques, which yields to gain efficiency, real-time adherence, repeatable mapping, and higher level of data privacy. This comes at the expense of information loss. However, this loss is controlled so that the data usability is not affected. In accordance with an embodiment, GT-ANeNDS can be applied to any data type for which a distance function can be defined. The higher level view of the algorithm is provided first and explained using numerical data type. In the discussion hereafter, by dataset we refer to a field, or a column, in the original database schema. FIG. 2 lists the main steps of the GT-ANeNDS approach. The input to the algorithm consists of the new transactional data item, and the meta-data. The meta-data consists of data-type, histogram and semantics:

• • [0035] Data-Type: The data-type is the regular database type, i.e., numerical, text, timestamp, etc. In addition to the semantics, datatype is used to determine the technique to use.
  • [0036] Histogram: The term histogram is used in a generic way to refer to the data structure that is incrementally maintained.
  • [0037] Semantics: The semantics of each data set is a record of the following information whenever applicable. Data-Sub-Type: for numerical data, the sub-type defines whether the data are general, or identifiable. Where identifiable data can identify the person, such as the national ID number. Euclidean distance Function: the function to be used to calculate the Euclidean distance between two values. The Origin point: the reference point of this data set.

[0038] Given the data-type and the semantics, the appropriate obfuscation technique is determined. In case it is GT-ANeNDS, the origin-point and the Euclidean distance function determine the appropriate bucket in the histogram, and the nearest neighbor therefore. Next, GT function is applied to the nearest neighbor, generating the obfuscated value. Next, we illustrate how the GT-ANeNDS works in case of numerical data types.

[0039] FIG. 3 illustrates histogram decomposition 132 for numerical data, in accordance with an embodiment. For general numerical data (i.e., non ID's such as bank account balance), the system can use equi-width histograms that splits the range of the data items distances into regions of the same width (i.e., range) to define the set of neighbors. Each bucket's 134 range is divided into a set of equi-height sub-buckets 136. The bucket's width and the subbucket's height are systems parameters set by the administrator. Histograms are built by scanning the current database shot once. The number of neighbors for each bucket depends on the height of the bucket 138 and the position of these neighbors depends on the values distribution in this range. Note that the horizontal axis is not the data value; however, it is the distance from the origin point. The vertical access is the frequency. This is introduced to be able to identify the nearest neighbor without the need to maintain any summary about the data values within each bucket.

[0040] The GT-ANeNDS process proceeds as follows. First, the distance between the original data item's value and the origin point is calculated, determining where in the histogram this data item falls. Second, the nearest neighbor point in the histogram is determined. The neighbors set, is the set of points determining sub-buckets' ranges within the same bucket this point belongs to. Finally, geometric transformation is applied to the nearest neighbor, generating the obfuscated value. A difference between the GT-NeNDS and GT-ANeNDS processes is that GT-ANeNDS uses a fixed set of neighbors for each bucket, which yields to map more than one original data value to the same obfuscated value, i.e., Anonymization. By fine tuning the bucket widths and the sub-bucket heights, the statistical characteristics of the original data are minimally impacted.

[0041] In accordance with an embodiment, for Boolean data-type, the same approach is used but the process simply uses two buckets only, and no sub-buckets. Therefore, the system can maintain in this case two counters for each bucket. To obfuscate a value, the new value is randomly drawn with probability to have the same ratio of the two values. For example, if it is a Gender field and the counters are: ten females and seven males, then the obfuscated value is set to M (i.e., male) with probability 7/17.

[0042] FIG. 4 illustrates a function for obfuscating identifiable numerical data 140, in accordance with an embodiment. For a numerical value is a key, such as national identification number, anonymization is not valid as it will result in distortion of the referential integrity constraints. In accordance with an embodiment, a Special Function 1 illustrated in FIG. 4 can be used. Opposed to NeNDS, the process can use a FaNDS technique (Farthest Neighbor Data Substitution). This is exactly same as NeNDS except that the process substitutes the data item with its farthest neighbor. Each digit of the original value is treated as a separate value to obfuscate. The set of digits are used as the neighbors from which the farthest neighbor is chosen to replace the original digit. Next, rotation is applied for each replaced digit and saved in a temporarily variable. This rotated number that results from replacing each digit in the original key and then rotating it is being added to the original key value and result is truncated to the key length and saved in a second temporarily variable. Finally, the obfuscated key is generated by randomly picking each digit from the two temporarily variables.

[0043] For date data type, neither GT-ANeNDS nor Special Function 1 fits, because of the semantics of the date. Therefore, in accordance with an embodiment the process can use a Special Function 2, to obfuscated date and timestamp data types, wherein the function basically utilizes controlled randomness to obfuscate each component of the date, i.e., the day, month and year.

[0044] FIG. 5 illustrates a table 144 of possible data-types and semantics, and which technique the system can use to obfuscate each data type, in accordance with an embodiment. In accordance with an embodiment, the system allows the user to overwrite these default selections and to define a user-defined obfuscation function. Depending on the implementation, the metadata about which technique to be used and its parameters can be stored in the original database itself, or in a parameters file.

[0045] In the following sections, the degree of data privacy, repeatability, and data usability of the proposed obfuscation techniques is analyzed.

[0046] Anonymization generally guarantees securing data 100%. Hence, numerical general data obfuscated using the GT-ANeNDS and that obfuscated using a dictionary are guaranteed to secure the privacy. For identifiable numerical data, Special Function 1 obfuscates the data using two different techniques then randomly picks digits from both obfuscated values into one new output value. Without full knowledge of the original data, there is no way to find out from where each digit was picked. Thus, data privacy is maintained, and the proposed obfuscation techniques are immune even to partial attacks, in which partial knowledge about the original data and/or the obfuscation process are used to reverse engineer a portion of the original data.

[0047] The proposed techniques guarantee obfuscation repeatability, i.e., applying to the same input data results in the same obfuscated data maintaining referential integrity. In the techniques used, the randomization can be dependant on the original data, i.e. the random seed is generated using the original data value, thus guaranteeing its repeatability.

[0048] Data usability is the hardest question to answer for numerical data since the proposed techniques introduce some anonymization. However, since the system determines the number of neighbors and their distances from the origin based on the number and distribution of data values within this bucket, thus the set of neighbors should be representative enough that the anonymized data are still useable.

[0049] In accordance with an embodiment, initial construction of the histograms and dictionaries is the only offline process within the system. Depending on the application dynamics, this process might need to be repeated, and the database rereplicated. This should be done in an efficient way, minimizing overhead and downtime.

[0050] In the following section, some performance results are described to provide a sense of how different techniques perform, and to demonstrate the data usability.

[0051] FIGS. 6 and 7 illustrate the results of the K-mean algorithm on original 150 and obfuscated data 152 respectively, in accordance with an embodiment. In a first experiment, the data usability of the system was demonstrated by applying K-mean classification algorithm, with k=8, using Weka Software to both the original and obfuscated data and plotting the results. The workload is a dataset of protein data in ARFF format. For the data obfuscation, the GT-ANeNDS was applied with theta equal to 45 degrees, origin point was set to the min value found in the original data set, and the histogram parameters were as follows: bucket width equals to one fourth of the range of the original data set, and sub-bucket height was set to 25%, so that there are four subbuckets in each bucket. As can be seen in FIGS. 6 and 7, the classification results are almost exactly the same, which demonstrates the data usability of the process.

[0052] FIG. 8 illustrates examples of original and obfuscated values, in accordance with an embodiment. In another experiment, an Oracle database was replicated to an MSSQL one using the system. One table was created that includes all different data types and obfuscated all fields except the notes, to identify the replicated record. The table shows the first five tuples, and their obfuscated replicas. As can be seen from the table, identifiable numerical values (SSN and credit card) are obfuscated using the Special Function 1 into unique (i.e., identifiable) values. The system also updated and deleted tuples as well, and the correct replica reflected the updates, showing the repeatability of the techniques. The table also shows for other data types how obfuscated values secure the original data.

[0053] The present invention may be conveniently implemented using one or more conventional general purpose or specialized digital computer, computing device, machine, or microprocessor, including one or more processors, memory and/or computer readable storage media programmed according to the teachings of the present disclosure. Appropriate software coding can readily be prepared by skilled programmers based on the teachings of the present disclosure, as will be apparent to those skilled in the software art.

[0054] In some embodiments, the present invention includes a computer program product which is a non-transitory storage medium or computer readable medium (media) having instructions stored thereon/in which can be used to program a computer to perform any of the processes of the present invention. The storage medium can include, but is not limited to, any type of disk including floppy disks, optical discs, DVD, CD-ROMs, microdrive, and magneto-optical disks, ROMs, RAMs, EPROMs, EEPROMs, DRAMs, VRAMs, flash memory devices, magnetic or optical cards, nanosystems (including molecular memory ICs), or any type of media or device suitable for storing instructions and/or data.

[0055] The foregoing description of the present invention has been provided for the purposes of illustration and description. It is not intended to be exhaustive or to limit the invention to the precise forms disclosed. Many modifications and variations will be apparent to the practitioner skilled in the art. The embodiments were chosen and described in order to best explain the principles of the invention and its practical application, thereby enabling others skilled in the art to understand the invention for various embodiments and with various modifications that are suited to the particular use contemplated. It is intended that the scope of the invention be defined by the following claims and their equivalence.