Automated Rule Generation for Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current rule-based anomaly detection systems in financial institutions are limited in their ability to detect new and unknown fraudulent activities due to reliance on static rule sets, requiring lengthy manual updates and expertise, and often generate complex, difficult-to-read rules that are inflexible and less effective than human-generated ones.
Innovation Solution
An automated rule generation system using Rule-Based-Machine Learning, which involves preparing a Key Indicator data warehouse and applying a custom decision tree machine learning algorithm to generate simple, effective rules for distinguishing fraudulent transactions, with rule evaluation and validation against business metrics.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual rule generation by business analysts is used, then rule accuracy and domain knowledge are improved, but time consumption and labor requirements increase significantly
Solution Approach 1:
The patent replaces the manual mechanical process of rule generation by business analysts with an automated machine learning system. The ML model processes transaction data and generates rules automatically, eliminating the time-consuming manual analysis while maintaining rule accuracy through data-driven patterns rather than human expertise alone.
Solution Approach 2:
The system enables self-service rule generation where the machine learning model autonomously analyzes transaction data, identifies patterns, and creates detection rules without requiring continuous human intervention. The system serves itself by automatically updating and optimizing rules based on new data, reducing dependency on business analyst expertise.
2Loss of time
If automated rule generation systems are used, then time consumption is reduced, but rule complexity and difficulty of understanding increase
Solution Approach 1:
The patent transforms the output parameters of the machine learning model into human-readable rule formats. Instead of presenting complex model predictions, the system converts them into simplified if-then rules with clear conditions and actions, making the automated generation process as interpretable as manual rule creation while maintaining the time efficiency of automation.
3Device complexity
If static rule sets are used, then system simplicity is maintained, but adaptability to new fraud patterns deteriorates
Solution Approach 1:
The patent introduces dynamic rule generation where the system continuously learns from new transaction data and automatically updates detection rules. Rather than using static rules that remain fixed, the system dynamically adapts its rule set based on emerging fraud patterns while maintaining overall system simplicity through automated management of the rule evolution process.
Data Source
AI summary
A processor is adapted to automatically generate and validate rules for monitoring suspicious activity by: For a first period of time, collecting a first group of transactions, automatically identifying and storing key indicators from the transactions, and automatically storing which of the transactions are pre-identified as fraudulent. Based on the key indicators and the pre-identified fraudulent transactions, training a learning algorithm and, with the learning algorithm, generating a decision tree of logical predicates including the key indicators. Based on the decision tree, generating a plurality of rules, each of which incorporates only one logical predicate from each layer of the decision tree. For a second period of time: collecting a second group of transactions, and generating a quality metric for each rule, by automatically testing the rules against the second group of transactions, and identifying a subset of rules for which the quality metric exceeds a threshold.