Fraud mitigation using one or more enhanced spatial features
Techniques are provided for fraud mitigation using enhanced spatial features. One method comprises obtaining transaction data associated with a transaction; obtaining a machine learning module trained using training transaction data for multiple geographic areas to learn a correlation of the training transaction data with fraudulent activity for each geographic area; extracting a transaction address from the transaction data; determining a given geographic area for the transaction using the transaction address; determining values for a predefined spatial feature for a predefined region that includes the transaction address in the given geographic area using a query of an external online data source; applying the determined values for the predefined spatial feature to the machine learning module to obtain an anomaly score for the transaction; and initiating a predefined remedial step and/or a predefined mitigation step when the transaction is determined to be a predefined anomaly based on the anomaly score.
1. A method, comprising:
obtaining transaction data associated with at least one transaction, wherein the transaction data comprises a transaction address comprising at least one of an order placement address, a shipping address and a billing address;
obtaining at least one machine learning module,
wherein the at least one machine learning module is trained using labeled training transaction data for a plurality of geographic areas to learn a correlation of at least a portion of the training transaction data with fraudulent activity for each geographic area;
in response to obtaining the transaction data associated with the at least one transaction:
extracting the transaction address from the transaction data;
determining a given geographic area of the plurality of geographic areas for the at least one transaction based at least in part on the transaction address;
querying at least one external online third-party data source, using a query comprising at least one query parameter derived from the transaction address, to determine values, by processing one or more results of the query, for one or more predefined spatial features for a predefined region that includes the transaction address in the given geographic area;
applying the determined values for the one or more predefined spatial features for the at least one transaction to the at least one machine learning module that generates an anomaly score indicating a likelihood that the at least one transaction is anomalous;
providing one or more of at least a portion of the transaction data, the anomaly score and a predefined anomaly label assigned to the at least one transaction in a feedback manner to update the at least one machine learning module; and
initiating one or more of a predefined automated remedial step and a predefined automated mitigation step in response to the at least one transaction being determined to be at least one of one or more predefined anomalies based at least in part on the anomaly score;
wherein the method is performed by at least one processing device comprising a processor coupled to a memory.
2. The method of claim 1 , wherein the at least one external online third-party data source comprises one or more of a municipal data source, a governmental data source and a private data source.
3. The method of claim 1 , wherein the one or more predefined anomalies comprise one or more of a risk anomaly and a fraud likelihood anomaly.
4. The method of claim 1 , wherein the at least one machine learning module is trained using training transaction data for a plurality of geographic areas labeled with one or more predefined anomaly labels.
5. The method of claim 1 , wherein the predefined region that includes the transaction address comprises a predefined area surrounding the transaction address.
6. The method of claim 1 , wherein the one or more predefined spatial features comprise a land use feature that determines a land use distribution indicating an anomaly likelihood for a plurality of land uses for at least some of the plurality of geographic areas.
7. The method of claim 1 , wherein the one or more predefined spatial features comprise a crime activity feature that determines a crime use distribution indicating an anomaly likelihood for a plurality of crimes for at least some of the plurality of geographic areas.
8. The method of claim 1 , wherein the one or more predefined spatial features comprise a port proximity feature that determines a port proximity distribution indicating an anomaly likelihood based at least in part on a distance between at least a portion of the transaction address and a shipping port for at least some of the plurality of geographic areas.
9. The method of claim 1 , wherein the one or more predefined spatial features comprise a census data feature that determines a census data distribution indicating an anomaly likelihood for a plurality of census data components for at least some of the plurality of geographic areas.
10. An apparatus comprising:
at least one processing device comprising a processor coupled to a memory;
the at least one processing device being configured to implement the following steps:
obtaining transaction data associated with at least one transaction, wherein the transaction data comprises a transaction address comprising at least one of an order placement address, a shipping address and a billing address;
obtaining at least one machine learning module,
wherein the at least one machine learning module is trained using labeled training transaction data for a plurality of geographic areas to learn a correlation of at least a portion of the training transaction data with fraudulent activity for each geographic area;
in response to obtaining the transaction data associated with the at least one transaction:
extracting the transaction address from the transaction data;
determining a given geographic area of the plurality of geographic areas for the at least one transaction based at least in part on the transaction address;
querying at least one external online third-party data source, using a query comprising at least one query parameter derived from the transaction address, to determine values, by processing one or more results of the query, for one or more predefined spatial features for a predefined region that includes the transaction address in the given geographic area;
applying the determined values for the one or more predefined spatial features for the at least one transaction to the at least one machine learning module that generates an anomaly score indicating a likelihood that the at least one transaction is anomalous;
providing one or more of at least a portion of the transaction data, the anomaly score and a predefined anomaly label assigned to the at least one transaction in a feedback manner for a retraining of the at least one machine learning module; and
initiating one or more of a predefined automated remedial step and a predefined automated mitigation step in response to the at least one transaction being determined to be at least one of one or more predefined anomalies based at least in part on the anomaly score.
11. The apparatus of claim 10 , wherein the one or more predefined spatial features comprise a land use feature that determines a land use distribution indicating an anomaly likelihood for a plurality of land uses for at least some of the plurality of geographic areas.
12. The apparatus of claim 10 , wherein the one or more predefined spatial features comprise a crime activity feature that determines a crime use distribution indicating an anomaly likelihood for a plurality of crimes for at least some of the plurality of geographic areas.
13. The apparatus of claim 10 , wherein the one or more predefined spatial features comprise a port proximity feature that determines a port proximity distribution indicating an anomaly likelihood based at least in part on a distance between at least a portion of the transaction address and a shipping port for at least some of the plurality of geographic areas.
14. The apparatus of claim 10 , wherein the one or more predefined spatial features comprise a census data feature that determines a census data distribution indicating an anomaly likelihood for a plurality of census data components for at least some of the plurality of geographic areas.
15. The apparatus of claim 10 , wherein the at least one machine learning module is trained using training transaction data for a plurality of geographic areas labeled with one or more predefined anomaly labels.
16. The apparatus of claim 10 , wherein the predefined region that includes the transaction address comprises a predefined area surrounding the transaction address.
17. A non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform the following steps:
obtaining transaction data associated with at least one transaction, wherein the transaction data comprises a transaction address comprising at least one of an order placement address, a shipping address and a billing address;
obtaining at least one machine learning module,
wherein the at least one machine learning module is trained using labeled training transaction data for a plurality of geographic areas to learn a correlation of at least a portion of the training transaction data with fraudulent activity for each geographic area;
in response to obtaining the transaction data associated with the at least one transaction:
extracting the transaction address from the transaction data;
determining a given geographic area of the plurality of geographic areas for the at least one transaction based at least in part on the transaction address;
querying at least one external online third-party data source, using a query comprising at least one query parameter derived from the transaction address, to determine values, by processing one or more results of the query, for one or more predefined spatial features for a predefined region that includes the transaction address in the given geographic area;
applying the determined values for the one or more predefined spatial features for the at least one transaction to the at least one machine learning module that generates an anomaly score indicating a likelihood that the at least one transaction is anomalous;
providing one or more of at least a portion of the transaction data, the anomaly score and a predefined anomaly label assigned to the at least one transaction in a feedback manner for a retraining of the at least one machine learning module; and
initiating one or more of a predefined automated remedial step and a predefined automated mitigation step in response to the at least one transaction being determined to be at least one of one or more predefined anomalies based at least in part on the anomaly score.
18. The non-transitory processor-readable storage medium of claim 17 , wherein the one or more predefined spatial features comprise one or more of a land use feature that determines a land use distribution indicating an anomaly likelihood for a plurality of land uses for at least some of the plurality of geographic areas and a crime activity feature that determines a crime use distribution indicating an anomaly likelihood for a plurality of crimes for at least some of the plurality of geographic areas.
19. The non-transitory processor-readable storage medium of claim 17 , wherein the one or more predefined spatial features comprise a port proximity feature that determines a port proximity distribution indicating an anomaly likelihood based at least in part on a distance between at least a portion of the transaction address and a shipping port for at least some of the plurality of geographic areas.
20. The non-transitory processor-readable storage medium of claim 17 , wherein the at least one machine learning module is trained using training transaction data for a plurality of geographic areas labeled with one or more predefined anomaly labels.