Artificial intelligence fraud management solution
An artificial intelligence fraud management solution comprises an expert programmer development system to build trainable general payment fraud models that integrate several artificial intelligence classifiers like neural networks, case based reasoning, decision trees, genetic algorithms, fuzzy logic, and rules and constraints. These are further integrated by the expert programmers and development system with smart agents and associated real-time profiling, recursive profiles, and long-term profiles. The trainable general payment fraud models are trained with supervised and unsupervised data to produce an applied payment fraud model. This then is applied by a commercial client to process real-time transactions and authorization requests for fraud scores.
1. A computer-implemented method for detecting payment fraud, comprising:
executing, at a computing device, an application including a classification model and a plurality of smart agents respectively corresponding to a plurality of transactional entities, each smart agent having a profile comprising a representation of historical data of the corresponding transactional entity and including long-term and real-time profiles with the real-time profile excluding historical data aged beyond a pre-determined time period, and the classification model being constructed according to one or more of: data mining logic, a neural network, case-based-reasoning, clustering, fuzzy logic, a genetic algorithm, a decision tree, and business rules;
inputting, via the computing device, transactional data to the application, the transactional data including transaction records reflecting transactions of the plurality of transactional entities, the application being configured to perform the following steps for each of the transaction records
compare the contents of at least one data field of the transaction record against each of the long-term profile and the real-time profile to generate a first output comprising an exception generated for each comparison result exceeding a corresponding threshold,
process the contents of at least one data field of the transaction record via the classification model to generate a second output,
analyze the first output and the second output using a weighted summation to generate a fraud score, including by adjusting the fraud score based on any generated exceptions,
wherein the application is further configured to
receive feedback to the fraud score comprising one of a false positive and a false negative indicator,
re-train the classification model based on the feedback,
update the long-term and real-time profiles based on the feedback,
process future transaction records using the re-trained classification model and the updated long-term and real-time profiles.
2. The computer-implemented method of claim 1 , wherein the classification model comprises data mining logic and the re-training comprises making an incremental change to a decision tree.
3. The computer-implemented method of claim 1 , wherein the classification model comprises a neural network and the re-training comprises updating a weight matrix of the neural network.
4. The computer-implemented method of claim 1 , wherein the application is further configured to update the profile of the corresponding smart agent to reflect contents of at least one data field of the transaction record.
5. The computer-implemented method of claim 1 , wherein analyzing the first output and the second output using the weighted summation includes retrieving one or more client tuning inputs and incorporating the one or more client tuning inputs into the weighted summation.
6. The computer-implemented method of claim 1 , wherein comparing the contents of the at least one data field of the transaction record against the real-time profile for each of the transaction records includes
determining whether revising the real-time profile to reflect the transaction record exceeds a threshold for transactions associated with the corresponding transactional entity within the pre-determined time period.
7. The computer-implemented method of claim 1 , wherein the profile of the corresponding smart agent includes a recursive profile.
8. The computer-implemented method of claim 7 , wherein
the application compares a plurality of data fields of the transaction record against the recursive profile,
an exception is generated for each recursive comparison result exceeding a corresponding threshold,
the weighted summation is configured to adjust the fraud score based on any generated recursive exceptions.
9. The computer-implemented method of claim 1 , wherein
the profile of the corresponding smart agent corresponds to a first transactional channel and to a first transactional entity of the plurality of transactional entities,
the application is further configured to
compare the contents of at least one data field of a second transaction record against a second profile corresponding to the first transactional entity and a second transactional channel to generate a third output,
process the contents of at least one data field of the second transaction record via a second classification model to generate a fourth output,
analyze the third output and the fourth output using one of the weighted summation and a second weighted summation to generate a second fraud score,
adjust the profile in response to at least one of the third output, the fourth output and the second fraud score,
adjust the second profile in response to at least one of the first output, the second output and the fraud score.
10. A computer program comprising instructions to perform the following steps, when executed by one or more processors:
compare contents of at least one data field of an instant transaction record against a corresponding profile of a smart agent, the profile including long-term and real-time profiles and comprising a representation of historical data of a corresponding transactional entity with the real-time profile excluding historical data aged beyond a pre-determined time period;
generate a first output from the comparison, the first output comprising an exception generated for each comparison result exceeding a corresponding threshold;
process contents of at least one data field of the instant transaction record via a classification model to generate a second output, the classification model being constructed according to one or more of: data mining logic, a neural network, case-based-reasoning, clustering, fuzzy logic, a genetic algorithm, a decision tree, and business rules;
analyse the first output and the second output using a weighted summation to generate a fraud score, including by adjusting the fraud score based on any generated exceptions;
receive feedback to the fraud score comprising one of a false positive and a false negative indicator;
re-train the classification model based on the feedback;
update the long-term and real-time profiles based on the feedback; and
process future transaction records using the re-trained classification model and the updated long-term and real-time profiles.
11. The computer program of claim 10 , wherein the classification model comprises data mining logic and the re-training comprises making an incremental change to a decision tree.
12. The computer program of claim 10 , wherein the classification model comprises a neural network and the re-training comprises updating a weight matrix of the neural network.
13. The computer program of claim 10 , wherein the program is further configured with instructions to update the profile of the corresponding smart agent to reflect contents of at least one data field of the transaction record.
14. The computer program of claim 10 , wherein analyzing the first output and the second output using the weighted summation includes retrieving one or more client tuning inputs and incorporating the one or more client tuning inputs into the weighted summation.
15. The computer program of claim 10 , wherein comparing the contents of the at least one data field of the transaction record against the real-time profile for each of the transaction records includes
determining whether revising the real-time profile to reflect the transaction record exceeds a threshold for transactions associated with the corresponding transactional entity within the pre-determined time period.
16. The computer program of claim 10 , wherein the profile of the corresponding smart agent includes a recursive profile.
17. The computer program of claim 16 , wherein
the program compares a plurality of data fields of the transaction record against the recursive profile,
an exception is generated for each recursive comparison result exceeding a corresponding threshold,
the weighted summation is configured to adjust the fraud score based on any generated exceptions.
18. The computer program of claim 10 , wherein
the profile of the corresponding smart agent corresponds to a first transactional channel and to the transactional entity,
the program is further configured with instructions to
compare the contents of at least one data field of a second transaction record against a second profile corresponding to the transactional entity and a second transactional channel to generate a third output,
process the contents of at least one data field of the second transaction record via a second classification model to generate a fourth output,
analyze the third output and the fourth output using one of the weighted summation and a second weighted summation to generate a second fraud score,
adjust the profile in response to at least one of the third output, the fourth output and the second fraud score,
adjust the second profile in response to at least one of the first output, the second output and the fraud score.