Machine learning for fraud tolerance
In an example embodiment, a solution is provided wherein a machine learning model is to determine a likelihood that a transaction is fraudulent, but also a separate machine learning model is used to determine a suitable threshold for a merchant. This predicted suitable threshold can either be automatically applied to the merchant, or can be recommended to the merchant (allowing the merchant to accept or reject it).
1. A method comprising:
accessing, by a device, historical data associated with historical network operations related to a set of computing infrastructures;
for each computing infrastructure of the set of computing infrastructures, extracting, by the device, one or more features from the historical data related to the computing infrastructure;
passing, by the device, the one or more extracted features to a second machine learning algorithm to train a set of instances of a second machine learning model corresponding to each computing infrastructure of the set of computing infrastructures,
each trained instance of the second machine learning model configured to output a determination of whether to allow or block a potential network operation based on (i) one or more features of the potential network operation, (ii) a decline score generated for the potential network operation by a first machine learning model, and (iii) a fraud tolerance threshold assigned to the computing infrastructure corresponding to the trained instance of the second machine learning model;
in response to obtaining a request for a first network operation with a first computing infrastructure, accessing, by the device, the first machine learning model trained by a first machine learning algorithm, the first machine learning model trained to output a first decline score indicative of a probability that a potential network operation is fraudulent;
executing, by the service, a first instance of the second machine learning model corresponding to the first computing infrastructure to evaluate the first network operation, based on the first decline score and a first fraud tolerance threshold assigned to the first computing infrastructure; and
in response to an indication from the first instance of the second machine learning model that the first network operation should be blocked, blocking, at the device, the first network operation such that the first network operation is prevented from proceeding externally outside of the device for further authentication processing.
2. The method of claim 1 , further comprising transforming the one or more features prior to them being passed to the second machine learning algorithm.
3. The method of claim 1 , further comprising:
measuring, by the device, results of the second machine learning algorithm; and
retraining, by the device, the second machine learning model using the measured results, wherein the measured results include a measure of how many of the network operations were disputed.
4. The method of claim 1 , wherein the second machine learning model is trained to assign a computing infrastructure to a segment that optimizes revenue of the computing infrastructure.
5. The method of claim 4 , wherein the segment is a grouping of at least a portion of the set of computing infrastructures who share a common fraud tolerance level.
6. The method of claim 1 , further comprising, while evaluating the potential network operation using the second machine learning model, dynamically adjusting the second machine learning model to effectively adjust the fraud tolerance threshold at runtime.
7. The method of claim 1 , further comprising adjusting the fraud tolerance threshold assigned to the computing infrastructure for the potential network operation based on a feature of the potential network operation.
8. A system comprising:
one or more processors; and
at least one memory storing instructions that, when executed by at least one processor among the one or more processors, cause the at least one processor to perform operations comprising:
accessing historical data associated with historical network operations related to a set of computing infrastructures;
for each computing infrastructure of the set of computing infrastructures, extracting one or more features from the historical data related to the computing infrastructure;
passing the one or more extracted features to a second machine learning algorithm to train a set of instances of a second machine learning model corresponding to each computing infrastructure of the set of computing infrastructures,
each trained instance of the second machine learning model configured to output a determination of whether to allow or block a potential network operation based on (i) one or more features of the potential network operation, (ii) a decline score generated for the potential network operation by a first machine learning model, and (iii) a fraud tolerance threshold assigned to the computing infrastructure corresponding to the trained instance of the second machine learning model;
in response to obtaining a request for a first network operation with a first computing infrastructure, accessing the first machine learning model trained by a first machine learning algorithm, the first machine learning model trained to output a first decline score indicative of a probability that a potential network operation is fraudulent;
executing a first instance of the second machine learning model corresponding to the first computing infrastructure to evaluate the first network operation based on the first decline score and a first fraud tolerance threshold assigned to the first computing infrastructure; and
in response to an indication from the first instance of the second machine learning model that the first network operation should be blocked, blocking the first network operation such that the first network operation is prevented from proceeding externally outside of the device for further authentication processing.
9. The system of claim 8 , wherein the instructions further cause the one or more processors to transform the one or more features prior to them being passed to the second machine learning algorithm.
10. The system of claim 8 , wherein the instructions further cause the one or more processor to:
measure results of the second machine learning algorithm; and
retrain the second machine learning model using the measured results, wherein the measured results include a measure of how many of the network operations were disputed.
11. The system of claim 8 , wherein the second machine learning model is trained to assign a computing infrastructure to a segment that optimizes revenue of the computing infrastructure.
12. The system of claim 11 , wherein the segment is a grouping of at least a portion of the set of computing infrastructures who share a common fraud tolerance level.
13. The system of claim 8 , further comprising, while evaluating the potential network operation using the second machine learning model, dynamically adjusting the second machine learning model to effectively adjust the fraud tolerance threshold at runtime.
14. The system of claim 8 , further comprising adjusting the fraud tolerance threshold assigned to the computing infrastructure for the potential network operation based on a feature of the potential network operation.
15. A non-transitory machine-readable medium comprising instructions which, when read by a machine, cause the machine to perform operations comprising:
accessing historical data associated with historical network operations related to a set of computing infrastructures;
for each computing infrastructure of the set of computing infrastructures, extracting one or more features from the historical data related to the computing infrastructure;
passing the one or more extracted features to a second machine learning algorithm to train a set of instances of a second machine learning model corresponding to each computing infrastructure of the set of computing infrastructures,
each trained instance of the second machine learning model configured to output a determination of whether to allow or block a potential network operation based on (i) one or more features of the potential network operation, (ii) a decline score generated for the potential network operation by a first machine learning model, and (iii) a fraud tolerance threshold assigned to the computing infrastructure corresponding to the trained instance of the second machine learning model;
in response to obtaining a request for a first network operation with a first computing infrastructure, accessing the first machine learning model trained by a first machine learning algorithm, the first machine learning model trained to output a first decline score indicative of a probability that a potential transaction is fraudulent;
executing a first instance of the second machine learning model corresponding to the first computing infrastructure to evaluate the first network operation based on the first decline score and a first fraud tolerance threshold assigned to the first computing infrastructure; and
in response to an indication from the first instance of the second machine learning model that the first network operation should be blocked, blocking the first network operation such that the first network operation is prevented from proceeding externally outside of the device for further authentication processing.
16. The non-transitory machine-readable medium of claim 15 , wherein the instructions further cause the machine to transform the one or more features prior to them being passed to the second machine learning algorithm.
17. The non-transitory machine-readable medium of claim 15 , wherein the instructions further cause the machine to:
measure results of the second machine learning algorithm; and
retrain the second machine learning model using the measured results, wherein the measured results include a measure of how many of the network operations were disputed.
18. The non-transitory machine-readable medium of claim 15 , wherein the second machine learning model is trained to assign a computing infrastructure to a segment that optimizes revenue of the computing infrastructure.
19. The non-transitory machine-readable medium of claim 15 , further comprising, while evaluating the potential network operation using the second machine learning model, dynamically adjusting the second machine learning model to effectively adjust the fraud tolerance threshold at runtime.
20. The non-transitory machine-readable medium of claim 15 , further comprising adjusting the fraud tolerance threshold assigned to the computing infrastructure for the potential network operation based on a feature of the potential network operation.