Automated transaction cancellation

1. A computing device comprising a processor and a memory configured to automatically suspend and schedule cancellation of fraudrisk transactions by executing a software application on a processor of the computing device to provide operations comprising:

receiving, from a database, a first dataset indicating a quantity of fraud-transactions, the first set of data generated at least in part by a fraud-score model;

receiving, from the database, a second dataset indicating a quantity of false positive fraud-transactions from the fraud-transactions;

transforming, by the processor, the first dataset and the second dataset into a fraud error rate;

comparing the fraud error rate with a predefined error rate;

generating, by the processor, in response to the error rate exceeding the predefined error rate, a new fraud-score model by:

constructing an initial modeling dataset of transactions, each having at least one transaction attribute, received from the database;

generating at least one bin from the initial modeling dataset to group at least a subset of the transactions having common transaction attributes; and

generating a modeling dataset based on the at least one bin and a statistical correlation to fraud-transactions;

determining a fraud cancellation model that includes the fraud-score model and a fraud-score cancel-threshold in response to a review of the new fraud-score model evaluated using pre-existing transaction data; and

using the new fraud cancellation model to evaluate a set of transactions and determine whether each of the transactions is a fraud-transaction.

2. The computing device of claim 1 , wherein the fraud-score cancel-threshold determined in response to the review is selected to achieve a specific objective related to at least one of a resource usage and a quality rate.

3. The computing device of claim 1 , further configured to execute the software application to provide operations comprising receiving additional first and second datasets and calculate a subsequent fraud error rate based on the additional datasets to determine whether to generate the new fraud-score model.

4. The computing device of claim 1 , wherein generating the new fraud-score model uses a set of fraudrisk transactions to generate the new fraud-score model, wherein the set of fraudrisk transactions is a subset of a total set of transactions.

5. The computing device of claim 1 , wherein generating the new fraud-score model includes generating multiple fraud-score models and selecting the best fraud-score model from among the multiple fraud-score models, the best fraud-score model being the fraud-score model that most accurately predicts fraud-transactions.

6. The computing device of claim 5 , wherein selecting the best fraud-score model comprises, when more than one model predicts fraud-transactions with the same accuracy, selecting the fraud-score model that identifies the most fraud-transactions with the same accuracy.

7. The computing device of claim 6 , further configured to execute the software application to provide operations comprising:

generating model performance information for the new fraud-score model, using pre-existing transaction data;

sending the model performance information for review; and

receiving a fraud-score cancel-threshold from the review.

8. A method for automatically suspending and scheduling cancellation of fraudrisk transactions comprising:

receiving, by a processor of a computing device from a database, a first dataset indicating a quantity of fraud-transactions, the first set of data generated at least in part by a fraud-score model;

receiving, by the processor of the computing device from the database, a second dataset indicating a quantity of false positive transactions from the fraud-transactions;

transforming, by the processor, the first dataset and the second dataset into a fraud error rate;

comparing, by the processor of the computing device, the fraud error rate with a predefined error rate; and

generating, by the processor of the computing device, in response to the error rate exceeding the predefined error rate, a new fraud-score model by:

constructing an initial modeling dataset of transactions, each having at least one transaction attribute, received from the database;

generating at least one bin from the initial modeling dataset to group at least a subset of the transactions having common transaction attributes; and

generating a modeling dataset based on the at least one bin and a statistical correlation to fraud-transactions;

wherein the second dataset indicating a quantity of false positive fraud-transactions includes fraud-transactions initially put on-hold and scheduled to be canceled where customer feedback identified the transaction as a false positive fraud-transaction before the transaction was canceled.

9. The method of claim 8 , further comprising:

determining a fraud cancellation model that includes the fraud-score model and a fraud-score cancel-threshold in response to the review of the model performance information generated from the new fraud-score model based at least in part on pre-existing transaction data; and

using the new fraud cancellation model to evaluate a set of transactions and determine whether each of the transactions is a fraud-transaction.

10. The method of claim 9 , wherein determining the fraud-score cancel-threshold in response to the review is selected to achieve a specific objective related to at least one of a resource usage and a quality rate.

11. The method of claim 10 , wherein the specific objective is to reduce the amount of fraud investigations by a predetermined amount.

12. The method of claim 8 , further comprising receiving additional first and second datasets at predetermined increments of time and calculating a subsequent fraud error rate based on the additional datasets to determine whether to generate the new fraud-score model.

13. The method of claim 8 , wherein generating the new fraud-score model includes generating multiple fraud-score models and selecting the best fraud-score model from among the multiple fraud-score models, the best fraud model being the fraud-score model that most accurately predicts fraud-transactions, and when more than one model predicts fraud-transactions with the same accuracy then selecting the fraud-score model that identifies the most fraud-transactions with the same accuracy.

14. A non-transitory computer-readable medium tangibly embodying computer-executable instructions comprising instructions that when executed by a processor of a computing device configured to automatically suspend and schedule cancellation of fraudrisk transactions causes the processor to:

receive, from a database, a first dataset indicating a quantity of fraud-transactions, the first set of data generated at least in part by a fraud-score model;

receive, from the database, a second dataset indicating a quantity of false positive fraud-transactions from the fraud-transactions;

compare the fraud error rate with a predefined error rate; and

generate, in response to the error rate exceeding the predefined error rate, a new fraud-score model by:

construct an initial modeling dataset of transactions, each having at least one transaction attribute, received from the database;

generate at least one bin from the initial modeling dataset to group at least a subset of the transactions having common transaction attributes; and

generate a modeling dataset based on the at least one bin and a statistical correlation to fraud-transactions;

wherein the generating of the new fraud-score model includes generating multiple fraud-score models and selecting the new fraud-score model from the multiple fraud-score models that most accurately predicts fraud-transactions, and when more than one model predicts fraud-transactions with the same accuracy then selecting the fraud-score model that identifies the most fraud-transactions with the same accuracy.

15. The computer-readable medium of claim 14 , wherein the second dataset indicating a quantity of false positive fraud-transactions includes fraud-transactions initially put on-hold and scheduled to be canceled where customer feedback identified the transaction as a false positive fraud-transaction before the transaction was canceled.

16. The computer-readable medium of claim 14 , further comprising instructions that when executed by a processor causes the processor to:

determine a fraud cancellation model that includes the fraud-score model and a fraud-score cancel-threshold in response to the review of the new fraud-score model evaluated using pre-existing transaction data; and

use the new fraud cancellation model to evaluate a set of transactions and determine whether each of the transactions is a fraud-transaction.

17. The computer-readable medium of claim 16 , wherein determining the fraud-score cancel-threshold in response to the review is selected to reduce the amount of fraud investigations by a predetermined amount.

18. The computer-readable medium of claim 14 , further comprising instructions that when executed by a processor causes the processor to receive additional first and second datasets and calculate a subsequent fraud error rate based on the additional datasets to determine whether to generate a new fraud-score model.