System, method, and computer program product for evaluating a fraud detection system

Patent metadata

Field	Value
Publication number	US-11244321-B2
Application number	US-201916590500-A
Country	US
Kind code	B2
Filing date	Oct 2, 2019
Priority date	Oct 2, 2019
Publication date	Feb 8, 2022
Grant date	Feb 8, 2022

Abstract

Provided are methods that include determining a set of transaction related actions for an agent, selecting a first transaction related action from the set of transaction related actions for the agent based on a plurality of features associated with the agent, generating transaction data associated with a fraudulent transaction based on the first transaction related action, generating a feature vector, the feature vector including transaction data associated with the fraudulent transaction, providing the feature vector as an input to a fraud detection machine learning model. Methods may also include determining an output of the fraud detection machine learning model based on the feature vector as the input, and generating a fraudulent reward parameter for the first transaction related action based on the output of the fraud detection machine learning model. Systems and computer program products are also provided.

First claim

What is claimed is:

1. A method, comprising:
determining, with at least one processor, a set of transaction related actions for an agent to be conducted during a simulation, wherein the agent comprises a simulated adversarial actor that is designed to engage in fraudulent conduct involving an account of a user, wherein each transaction related action comprises an action associated with conducting a payment transaction by the agent;
selecting, with at least one processor, a first transaction related action from the set of transaction related actions for the agent based on a plurality of features associated with the agent, wherein selecting the first transaction related action from the set of transaction related actions for the agent comprises:
determining the plurality of features associated with the agent based on one or more historical payment transactions of a plurality of historical payment transactions involving the account of the user; and
selecting the first transaction related action from the set of transaction related actions for the agent based on an output from an agent action machine learning model and the plurality of features associated with the agent, wherein the plurality of features associated with the agent are provided as an input to the agent action machine learning model and the first transaction related action is an output of the agent action machine learning model based on the input;

generating, with at least one processor, transaction data associated with a fraudulent transaction based on the first transaction related action;
generating, with at least one processor, a feature vector, wherein the feature vector comprises transaction data associated with the fraudulent transaction;
providing, with at least one processor, the feature vector as an input to a fraud detection machine learning model;
determining, with at least one processor, an output of the fraud detection machine learning model based on the feature vector as the input;
generating, with at least one processor, a fraudulent reward parameter for the first transaction related action based on the output of the fraud detection machine learning model; and
updating, with at least one processor, a weight parameter of the agent action machine learning model based on the fraudulent reward parameter and a reinforcement learning algorithm, the reinforcement learning algorithm being defined as follows:
R t =r t+1 +γr t+2 +γ 2 r t+3 +γ 3 r t+4 + . . . =Σ k=0 ∞ γ k r t+k+1 ,

where r t+k+1 is a predicted fraudulent reward amount for a time interval of the simulation, wherein the predicted fraudulent reward amount is associated with a fraudulent payment transaction that will be conducted by the agent, and y k is a weight parameter with value range: 0<γ<1 that is chosen to optimize a fraudulent reward amount parameter, R t .

2. The method of claim 1 , further comprising:
generating, with at least one processor, a plurality of status indicators regarding fraudulent transaction outcomes associated with transaction related actions performed by the agent based on transaction data associated with the plurality of historical payment transactions, wherein each status indicator comprises an indication of a status of each historical payment transaction as being a fraudulent transaction or a non-fraudulent transaction, and
wherein generating the feature vector comprises:
generating, with at least one processor, the feature vector based on the transaction data associated with the plurality of historical payment transactions, wherein the transaction data associated with the plurality of historical payment transactions comprises the plurality of status indicators, wherein the feature vector comprises transaction data associated with the fraudulent transaction, and wherein the fraudulent transaction is a fraudulent transaction of the plurality of historical payment transactions.

3. The method of claim 1 , wherein determining the set of transaction related actions for the agent comprises: determining the set of transaction related actions for the agent based on historical transaction data associated with the one or more historical transactions.

4. The method of claim 1 , further comprising:
updating the agent action machine learning model based on a fraudulent reward parameter for another transaction related action.

5. The method of claim 1 , further comprising:
assigning, with at least one processor, the fraudulent reward parameter to the first transaction related action; and
storing, with at least one processor, the first transaction related action and the fraudulent reward parameter in a data structure.

6. The method of claim 1 , further comprising:
generating, with at least one processor, a plurality of fraudulent reward parameters for a sequence of transaction related actions based on a plurality of outputs of the fraud detection machine learning model;
determining, with at least one processor, a fraudulent reward amount based on the plurality of fraudulent reward parameters;
assigning, with at least one processor, the fraudulent reward amount to the sequence of transaction related actions; and
storing the fraudulent reward amount and the sequence of transaction related actions in a data structure.

7. The method of claim 6 , further comprising:
determining, with at least one processor, a plurality of sequences of transaction related actions that are each associated with a fraudulent reward amount; and
selecting, with at least one processor, a sequence of transaction related actions from the plurality of sequences of transaction related actions that is associated with a maximum reward amount of a plurality of fraudulent reward amounts.

8. The method of claim 1 , further comprising:
determining, with at least one processor, whether the fraud detection machine learning model is deployed in an active setting; and
performing, with at least one processor, an action associated with enhancing fraud detection for a transaction to be conducted based on determining that the fraud detection machine learning model is deployed in the active setting.

9. A system, comprising:
at least one processor programmed or configured to:
determine a set of transaction related actions for an agent to be conducted during a simulation, wherein the agent comprises a simulated adversarial actor that is designed to engage in fraudulent conduct involving an account of a user, wherein each transaction related action comprises an action associated with conducting a payment transaction by the agent;
select a first transaction related action from the set of transaction related actions for the agent based on a plurality of features associated with the agent; wherein, when selecting the first transaction related action, the at least one processor is programmed or configured to:
determine the plurality of features associated with the agent based on one or more historical payment transactions of a plurality of historical payment transactions involving the account of the user;
select the first transaction related action from the set of transaction related actions for the agent based on an output from an agent action machine learning model and the plurality of features associated with the agent, wherein the plurality of features associated with the agent are provided as an input to an agent action machine learning model, and the first transaction related action is the output of the agent action machine learning model based on the input;

generate transaction data associated with a fraudulent transaction based on the first transaction related action;
generate a feature vector, wherein the feature vector comprises transaction data associated with the fraudu

Assignees

• Visa Int Service Ass

Inventors

• Wu Yuhang
• Yang Hao

Classifications

• G06F16/22

  Indexing; Data structures therefor; Storage structures · CPC title

• G06N20/00

  Machine learning · CPC title

• G06Q20/4016 Primary

  involving fraud or risk level assessment in transaction processing · CPC title