Unsupervised Feature Selection for Low Fraud ML Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Financial institutions face challenges in creating robust machine learning models for fraud detection due to low fraud counts in transaction data sets, leading to inadequate training data and increased false positives, which can result in significant financial losses.
Innovation Solution
A hybrid machine learning model is generated using transfer learning for feature selection, where SHapley Additive exPlanations (SHAP) are used to rank features across different financial institutions, and an automated script identifies a robust subset of features for improved fraud detection in low fraud scenarios.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a simple linear regression model is created due to low fraud counts, then the model can be trained with limited data, but the model fails to identify true fraud and increases false positives
Solution Approach 1:
The patent combines multiple financial institutions' transaction data sets through a federated learning framework, merging data from multiple sources to create a more robust training data set while maintaining data privacy through secure multi-party computation
Solution Approach 2:
The patent performs preliminary feature selection using SHAP (SHapley Additive exPlanations) values to identify and select the most important features before model training, ensuring that the model trains on high-quality, relevant features even when data is limited
2Ease of operation
If manual feature selection by subject matter experts is used, then the model can be re-trained with selected features, but it is not possible to cover all fraud detection scenarios and select the best features from different tenants
Solution Approach 1:
The patent implements automated feature selection using SHAP values that automatically identify and rank features based on their importance to fraud detection, eliminating the need for manual expert selection and enabling the system to adapt to different fraud scenarios autonomously
Solution Approach 2:
The patent creates a universal feature selection mechanism that works across multiple financial institutions and different fraud scenarios, allowing the same automated SHAP-based approach to identify relevant features for various types of fraud detection problems
3Adaptability or versatility
If transfer learning is used for feature selection across different tenants, then the best features can be selected dynamically, but the system complexity increases
Solution Approach 1:
The patent introduces an intermediary layer that computes SHAP values to bridge the gap between raw transaction data and model predictions, enabling automated feature selection without requiring complex custom implementations for each institution
Data Source
AI summary
A machine learning (ML) system configured to detect fraud in tenant data systems. The system includes a processor and a computer readable medium operably coupled thereto, the computer readable medium comprising a plurality of instructions stored in association therewith that are accessible to, and executable by, the processor, to perform ML modeling operations which include receiving a first data set, determining that the first data set meets or exceeds a low fraud tenant threshold, segmenting the first tenant data system based on the first data set, determining first features of a first ML model, determining a first explanation of a first feature importance of each of the first features, comparing the first tenant data system to a second tenant data system based on at least the first explanation and a second explanation, ranking at least the first features and the second features, and performing a feature selection.