Fraud Rule Generation via Adaptive Thresholding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Fraud detection systems face challenges in adapting to evolving fraudulent tactics, as fraudsters continually change their methods, making it difficult to effectively identify and flag fraudulent behavior.
Innovation Solution
A computer system and method that generates fraud rule criteria by creating and categorizing training data, calculating metrics, and selecting cutoff values to flag risky groups, which are then used to generate fraud rule criteria, allowing for continuous adaptation to new fraud patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If traditional fraud detection algorithms are used, then existing fraud patterns can be detected, but the system cannot adapt to evolving fraudulent tactics
Solution Approach 1:
The system performs preliminary actions by continuously training on recent transaction data before new fraud patterns fully emerge. This proactive training approach allows the model to learn and adapt to evolving fraud tactics ahead of time, improving detectability of new patterns while maintaining reliability through continuous validation
Solution Approach 2:
The system implements feedback loops where detection results and new transaction data continuously feed back into model retraining. This closed-loop approach allows the system to learn from both confirmed fraud cases and legitimate transactions, continuously improving adaptability to new tactics while maintaining detection accuracy through iterative optimization
2Reliability
If fraud detection rules are updated frequently to catch new tactics, then detection accuracy improves, but system complexity increases
Solution Approach 1:
The system performs self-service by automatically retraining and updating its own fraud detection models without requiring manual intervention for each rule update. This automation reduces system complexity by eliminating the need for manual rule management while maintaining high detection accuracy through continuous self-optimization
Solution Approach 2:
The system manages complexity by changing parameters incrementally through continuous training on new data rather than making large, discrete rule updates. This gradual parameter adjustment allows the model to adapt to new fraud tactics while maintaining system stability and reducing operational complexity
3Reliability
If more training data is used to improve model accuracy, then detection performance improves, but processing time increases
Solution Approach 1:
The system performs preliminary training actions on batches of recent transaction data before they are needed for detection. This advance processing allows the model to be pre-adapted to emerging patterns, improving detection performance while minimizing real-time processing delays
Solution Approach 2:
The system implements periodic training cycles where models are retrained at scheduled intervals or when certain data thresholds are met. This periodic approach balances the need for up-to-date detection performance with acceptable processing times by concentrating computational effort in periodic batches rather than continuous real-time processing
Data Source
AI summary
A computer system comprises at least one processor; and a memory coupled to the at least one processor and storing processor-executable instructions which, when executed by the at least one processor, configure the at least one processor to create a first set of training data that includes data flagged as fraud and data flagged as not fraud; categorize the first set of training data into a number of first groups; for each first group, calculate at least one metric; compare the at least one metric to a number of first cutoff values; select a first cutoff value that generates a maximum performance output as a first threshold; flag at least one first group that has the at least one metric below the first threshold as risky; and generate fraud rule criteria based on the at least one first group.