Reducing false positives using customer feedback and machine learning
A method of reducing a future amount of electronic fraud alerts includes receiving data detailing a financial transaction, inputting the data into a rules-based engine that generates an electronic fraud alert, transmitting the alert to a mobile device of a customer, and receiving from the mobile device customer feedback indicating that the alert was a false positive or otherwise erroneous. The method also includes inputting the data detailing the financial transaction into a machine learning program trained to (i) determine a reason why the false positive was generated, and (ii) then modify the rules-based engine to account for the reason why the false positive was generated, and to no longer generate electronic fraud alerts based upon (a) fact patterns similar to fact patterns of the financial transaction, or (b) data similar to the data detailing the financial transaction, to facilitate reducing an amount of future false positive fraud alerts.
1. A computer-implemented method of predicting a fraud classification using fraud classification rules, comprising:
training, by a processor, a machine learning program using training data labeled with fraud classification labels, wherein:
the fraud classification labels identify types of fraud, of a set of different predetermined types of fraud, associated with respective historical transactions;
generating, by the processor, the fraud classification rules using the trained machine learning program; and
predicting, by the processor, and by applying the fraud classification rules to account data associated with a particular financial account, the fraud classification that:
corresponds to a transaction associated with the particular financial account, and
identifies a particular type of fraud, included in the set of different predetermined types of fraud, that is predicted to be associated with the transaction.
2. The computer-implemented method of claim 1 , further comprising:
receiving, by the processor, feedback indicating an accuracy of the fraud classification;
re-training, by the processor, the machine learning program using the feedback; and
updating, by the processor, the fraud classification rules based on the re-training of the machine learning program.
3. The computer-implemented method of claim 1 , wherein training the machine learning program comprises identifying factors, indicated by the training data, that correlate to the types of fraud, associated with the respective historical transactions, identified by the fraud classification labels.
4. The computer-implemented method of claim 3 , further comprising identifying, by the processor, different subsets of the factors that are respectively associated with the types of fraud.
5. The computer-implemented method of claim 3 , further comprising:
determining, by the processor, weights corresponding to the factors associated with the particular type of fraud; and
predicting, by the processor, the fraud classification based at least in part on the weights and instances of the factors that are indicated by the account data.
6. The computer-implemented method of claim 5 , wherein predicting the fraud classification comprises:
determining, by the processor, and based on the weights and the instances of the factors, a fraud classification score associated with the particular type of fraud; and
determining, by the processor, that the fraud classification score exceeds a threshold score.
7. The computer-implemented method of claim 1 , wherein predicting the fraud classification comprises:
predicting, by the processor, a first fraud classification associated with a first type of fraud, the first fraud classification being associated with a first fraud classification score;
predicting, by the processor, a second fraud classification associated with a second type of fraud, the second fraud classification being associated with a second fraud classification score; and
determining, by the processor, the particular type of fraud as one of the first type of fraud or the second type of fraud based on a higher score of the first fraud classification score and the second fraud classification score.
8. The computer-implemented method of claim 1 , wherein the set of different predetermined types of fraud comprises two or more of: counterfeiting, forgery, account takeover, lost card use, stolen card use, skimming, chargeback fraud, application fraud, or a lack of fraud.
9. The computer-implemented method of claim 1 , wherein the fraud classification is predicted in response to an indication of suspect activity associated with at least one of the transaction or the particular financial account.
10. A computer system configured to predict a fraud classification using fraud classification rules, the computer system comprising:
a processor; and
memory storing computer-executable instructions that, when executed by the processor, cause the computer system to:
train a machine learning program using training data labeled with fraud classification labels, wherein:
the fraud classification labels identify types of fraud, of a set of different predetermined types of fraud, associated with respective historical transactions;
generate, using the trained machine learning program, the fraud classification rules; and
predict, by applying the fraud classification rules to account data associated with a particular financial account, the fraud classification that:
corresponds to a transaction associated with the particular financial account, and
identifies a particular type of fraud, included in the set of different predetermined types of fraud, that is predicted to be associated with the transaction.
11. The computer system of claim 10 , wherein the computer-executable instructions further cause the computer system to:
receive feedback indicating an accuracy of the fraud classification;
re-train the machine learning program using the feedback; and
update the fraud classification rules based on re-training of the machine learning program.
12. The computer system of claim 10 , wherein the computer-executable instructions cause the computer system to train the machine learning program by identifying factors, indicated by the training data, that correlate to the types of fraud, associated with the respective historical transactions, identified by the fraud classification labels.
13. The computer system of claim 12 , wherein the computer-executable instructions cause the computer system to:
determine weights corresponding to the factors associated with the particular type of fraud; and
predict the fraud classification based at least in part on the weights and instances of the factors that are indicated by the account data.
14. The computer system of claim 10 , wherein the computer-executable instructions cause the computer system to predict the fraud classification by:
predicting a first fraud classification associated with a first type of fraud, the first fraud classification being associated with a first fraud classification score;
predicting a second fraud classification associated with a second type of fraud, the second fraud classification being associated with a second fraud classification score; and
determining the particular type of fraud as one of the first type of fraud or the second type of fraud based on a higher score of the first fraud classification score and the second fraud classification score.
15. The computer system of claim 10 , wherein the computer-executable instructions cause the computer system to predict the fraud classification in response to an indication of suspect activity associated with at least one of the transaction or the particular financial account.
16. One or more non-transitory computer-readable media storing computer-executable instructions for using fraud classification rules to predict a fraud classification that, when executed by a processor, cause the processor to:
train a machine learning program using training data labeled with fraud classification labels, wherein:
the fraud classification labels identify types of fraud, of a set of different predetermined types of fraud, associated with respective historical transactions;
generate, using the trained machine learning program, the fraud classification rules; and
predict, by applying the fraud classification rules to account data associated with a particular financial account, the fraud classification that:
corresponds to a transaction associated with the particular financial account, and
identifies a particular type of fraud, included in the set of different predetermined types of fraud, that is predicted to be associated with the transaction.
17. The one or more non-transitory computer-readable media of claim 16 , wherein the computer-executable instructions further cause the processor to:
receive feedback indicating an accuracy of the fraud classification;
re-train the machine learning program using the feedback; and
update the fraud classification rules based on re-training of the machine learning program.
18. The one or more non-transitory computer-readable media of claim 16 , wherein the computer-executable instructions cause the processor to train the machine learning program by identifying factors, indicated by the training data, that correlate to the types of fraud, associated with the respective historical transactions, identified by the fraud classification labels.
19. The one or more non-transitory computer-readable media of claim 16 , wherein the computer-executable instructions cause the processor to predict the fraud classification by:
predicting a first fraud classification associated with a first type of fraud, the first fraud classification being associated with a first fraud classification score;
predicting a second fraud classification associated with a second type of fraud, the second fraud classification being associated with a second fraud classification score; and
determining the particular type of fraud as one of the first type of fraud or the second type of fraud based on a higher score of the first fraud classification score and the second fraud classification score.
20. The one or more non-transitory computer-readable media of claim 16 , wherein the computer-executable instructions cause the processor to predict the fraud classification in response to an indication of suspect activity associated with at least one of the transaction or the particular financial account.