US5790645A
Expired / Lapsed
US
⏱️ 0.0 years remaining
Automatic design of fraud detection systems
Assignee: Verizon Patent and Licensing Inc
•
Filing: 1996-08-01
•
Granted: 1998-08-04
•
Expires: 2016-08-01
Abstract
Abstract A technique for automatically designing a fraud detection system using a series of machine learning methods. Data mining and constructive induction are combined with more standard machine learning techniques to design methods for detecting fraudulent usage based on profiling customer behavior. Specifically, a rule-learning is used to uncover indicators of fraudulent behavior from a large user database. These indicators are used to create profilers, which then serve as features to the fraud detection system that combines evidence from multiple profilers to generate high-confidence intervention activities when the system is deployed on-line with user data.
Inventors
- 👤 Tom Elliott Fawcett
- 👤 Foster John Provost
Bibliographic Details
Application Number
—
Jurisdiction
US
Data Source
google_patents
Last Synchronized
2026-10-27 09:32
CPC Classifications
H; H04; H04W; H04W12/00; H04W12/12; H04W12/126; G; G06; G06F; G06F18/00; G06F18/20; G06F18/24; G06F18/24765
Claims Analysis
Total Claims: 26
Claim 1
Independent Claim
102 words
A method for automatically generating a fraud detection system from recent user information wherein individual user information is summarized by specified indicators, the method comprising the steps of supplying rules indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, instantiating profilers with reference to the user information and the rules, each of the profilers characterizing the behavior of the user with respect to one of the subset of the indicators, and learning to combine outputs from the profilers utilizing the user information to produce a trained set of profilers representative of the fraud detection system.
Claim 2
Independent Claim
102 words
A method for automatically generating a fraud detection system from recent user information wherein individual user information is summarized by specified indicators, the method comprising the steps of supplying rules indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, instantiating profilers with reference to the user information and the rules, each of the profilers characterizing the behavior of the user with respect to one of the subset of the indicators, and learning to combine outputs from the profilers utilizing the user information to produce a trained set of profilers representative of the fraud detection system.
Claim 3
Dependent Claim
23 words
The method as recited in claim 1 wherein the step of learning to combine outputs includes the step of optimizing the learning step.
Claim 4
Dependent Claim
51 words
The method as recited in claim 1 wherein the trained set of profilers is characterized by a set of weights to weight the outputs of the profilers and by a comparison threshold, and the step of learning includes the step of determining the set of weights, or the threshold, or both.
Claim 5
Dependent Claim
43 words
The method as recited in claim 1 wherein the method is carried out on a programmed digital computer and further including the step of storing the output of the learning step in the computer as program code representative of the fraud detection system.
Claim 6
Independent Claim
127 words
A method for automatically generating a fraud detection system from recent user information wherein individual user information is summarized by specified indicators, the fraud detection system including a set of trained profilers and an evaluator which combines the outputs of the trained profilers for comparison to a comparison threshold, the method comprising the steps of supplying rules indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, instantiating profilers with reference to the user information and the rules, each of the profilers characterizing the behavior of the user with respect to one of the subset of the indicators, and learning to combine outputs from the profilers utilizing the user information to produce the trained set of profilers and the comparison threshold.
Claim 7
Independent Claim
127 words
A method for automatically generating a fraud detection system from recent user information wherein individual user information is summarized by specified indicators, the fraud detection system including a set of trained profilers and an evaluator which combines the outputs of the trained profilers for comparison to a comparison threshold, the method comprising the steps of supplying rules indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, instantiating profilers with reference to the user information and the rules, each of the profilers characterizing the behavior of the user with respect to one of the subset of the indicators, and learning to combine outputs from the profilers utilizing the user information to produce the trained set of profilers and the comparison threshold.
Claim 8
Dependent Claim
23 words
The method as recited in claim 5 wherein the step of learning to combine outputs includes the step of optimizing the learning step.
Claim 9
Dependent Claim
46 words
The method as recited in claim 5 wherein the trained set of profilers is characterized by a set of weights to weight the outputs of the profilers, and the step of learning includes the step of determining the set of weights, or the threshold, or both.
Claim 10
Dependent Claim
43 words
The method as recited in claim 5 wherein the method is carried out on a programmed digital computer and further including the step of storing the output of the learning step in the computer as program code representative of the fraud detection system.
Claim 11
Independent Claim
206 words
A method for automatically generating a fraud detection system from recent user information wherein individual user information is summarized by specified indicators, the fraud detection system being arranged to process incoming user information, the method comprising the steps of supplying rules indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, instantiating profilers with reference to the recent user information and the rules, each of the profilers characterizing the behavior of the user with respect to one of the subset of the indicators, and learning to combine outputs from the profilers utilizing the recent user information to produce a trained set of profilers representative of the fraud detection system, wherein the trained set of profilers is characterized by a set of weights to weight the outputs of the profilers and a comparison threshold, and the step of learning includes the step of determining the set of weights, or the threshold, or both, and wherein to implement the fraud detection system, the profilers process the incoming user information to produce processed outputs, the processed outputs are weighted by the respective weights to produce weighted outputs, and the weighted outputs are combined for comparison with the comparison threshold to determine fraudulent activity.
Claim 12
Independent Claim
206 words
A method for automatically generating a fraud detection system from recent user information wherein individual user information is summarized by specified indicators, the fraud detection system being arranged to process incoming user information, the method comprising the steps of supplying rules indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, instantiating profilers with reference to the recent user information and the rules, each of the profilers characterizing the behavior of the user with respect to one of the subset of the indicators, and learning to combine outputs from the profilers utilizing the recent user information to produce a trained set of profilers representative of the fraud detection system, wherein the trained set of profilers is characterized by a set of weights to weight the outputs of the profilers and a comparison threshold, and the step of learning includes the step of determining the set of weights, or the threshold, or both, and wherein to implement the fraud detection system, the profilers process the incoming user information to produce processed outputs, the processed outputs are weighted by the respective weights to produce weighted outputs, and the weighted outputs are combined for comparison with the comparison threshold to determine fraudulent activity.
Claim 13
Dependent Claim
23 words
The method as recited in claim 9 wherein the step of learning to combine outputs includes the step of optimizing the learning step.
Claim 14
Dependent Claim
43 words
The method as recited in claim 9 wherein the method is carried out on a programmed digital computer and further including the step of storing the output of the learning step in the computer as program code representative of the fraud detection system.
Claim 15
Independent Claim
103 words
A method for automatically generating a telephony fraud detection system from recent call data wherein individual call data is summarized by specified indicators, the method comprising the steps of supplying rules indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, instantiating profilers with reference to the call data and the rules, each of the profilers characterizing the behavior of the user with respect to one of the subset of the indicators, and learning to combine outputs from the profilers utilizing the call data to produce a trained set of profilers representative of the fraud detection system.
Claim 16
Independent Claim
103 words
A method for automatically generating a telephony fraud detection system from recent call data wherein individual call data is summarized by specified indicators, the method comprising the steps of supplying rules indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, instantiating profilers with reference to the call data and the rules, each of the profilers characterizing the behavior of the user with respect to one of the subset of the indicators, and learning to combine outputs from the profilers utilizing the call data to produce a trained set of profilers representative of the fraud detection system.
Claim 17
Dependent Claim
23 words
The method as recited in claim 12 wherein the step of learning to combine outputs includes the step of optimizing the learning step.
Claim 18
Dependent Claim
51 words
The method as recited in claim 12 wherein the trained set of profilers is characterized by a set of weights to weight the outputs of the profilers and by a comparison threshold, and the step of learning includes the step of determining the set of weights, or the threshold, or both.
Claim 19
Dependent Claim
43 words
The method as recited in claim 12 wherein the method is carried out on a programmed digital computer and further including the step of storing the output of the learning step in the computer as program code representative of the fraud detection system.
Claim 20
Independent Claim
126 words
A method for automatically generating a fraud detection system from recent user information wherein individual user information is described by specified indicators, the method comprising the steps of generating rules by processing the user information, the rules being indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, defining profiler templates, each of the profiler templates corresponding to a type of profiler, each profiler characterizing the behavior of the user with respect to one of the subset of the indicators, instantiating profilers with reference to the user information, the rules, and the profiler templates, and learning to combine outputs from the profilers utilizing the user information to produce a trained set of profilers for use as the fraud detection system.
Claim 21
Independent Claim
126 words
A method for automatically generating a fraud detection system from recent user information wherein individual user information is described by specified indicators, the method comprising the steps of generating rules by processing the user information, the rules being indicative of a subset of the indicators useful in distinguishing legitimate user behavior from fraudulent behavior, defining profiler templates, each of the profiler templates corresponding to a type of profiler, each profiler characterizing the behavior of the user with respect to one of the subset of the indicators, instantiating profilers with reference to the user information, the rules, and the profiler templates, and learning to combine outputs from the profilers utilizing the user information to produce a trained set of profilers for use as the fraud detection system.
Claim 22
Dependent Claim
23 words
The method as recited in claim 16 wherein the step of learning to combine outputs includes the step of optimizing the learning step.
Claim 23
Dependent Claim
78 words
The method as recited in claim 16 wherein each of the templates includes a rule field, a training field, and a use field, and wherein said step of defining profiler templates includes the steps of defining, for each of the profiler templates, a training operation on the user information for a given one of the rules listed in the rule field, and defining a use operation for processing the user information for the given one of the rules.
Claim 24
Dependent Claim
21 words
The method as recited in claim 16 wherein the step of generating includes the step of data mining the user information.
Claim 25
Dependent Claim
98 words
The method as recited in claim 16 wherein the trained set of profilers is characterized by a set of weights to weight the outputs of the profilers and a comparison threshold, and the step of learning includes the step of determining the set of weights, or the threshold, or both, and wherein to implement the fraud detection system, the profilers process the incoming user information to produce processed outputs, the processed outputs are weighted by the respective weights to produce weighted outputs, and the weighted outputs are combined for comparison with the comparison threshold to determine fraudulent activity.
Claim 26
Dependent Claim
43 words
The method as recited in claim 16 wherein the method is carried out on a programmed digital computer and further including the step of storing the output of the learning step in the computer as program code representative of the fraud detection system.
References Cited (Backward Citations)
USPTO Annuity & Maintenance Windows
| Annuity Stage | Window Opens | Payment Deadline | Surcharge Period Until | Estimated Fee | Status |
|---|---|---|---|---|---|
| 3.5 Years (Year 4) | 2001-08-03 | 2002-02-02 | 2002-08-04 | $2000.00 | Window Closed / Lapsed |
| 7.5 Years (Year 8) | 2005-08-03 | 2006-02-02 | 2006-08-04 | $3760.00 | Window Closed / Lapsed |
| 11.5 Years (Year 12) | 2009-08-03 | 2010-02-02 | 2010-08-04 | $7700.00 | Window Closed / Lapsed |