Merchant Data Shift Detection via Time Series and Isolation Forest
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current risk scoring models and fraud detection systems in electronic payment networks are inefficient and inaccurate in detecting high-risk behavior merchants due to dynamic merchant behaviors, high dimensionality, and seasonality trends in transaction data, particularly in identifying merchant data shifts across Merchant Category Codes (MCCs).
Innovation Solution
A computer-implemented method using a combination of time series analysis and machine learning, specifically applying a difference transform and training a machine learning model with an isolation forest algorithm to generate anomaly scores, which identifies shifts in transaction volumes across MCCs by comparing historical and previous transaction data to detect potential fraud or MCC gaming.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional risk scoring models are used to detect high-risk merchant behavior, then the authorization process can be maintained with standard procedures, but the detection accuracy and efficiency of fraudulent transactions remain insufficient
Solution Approach 1:
The patent segments the transaction data analysis by dividing it into multiple MCC categories and time series periods. The system separately analyzes transaction patterns across different MCCs and time periods, then compares them to detect anomalies. This segmentation enables more precise detection of merchant behavior changes without overwhelming the system with aggregated data, thereby improving detection accuracy while maintaining processing efficiency.
Solution Approach 2:
The patent introduces a new dimension of analysis by applying time series transformation to transaction data. Instead of analyzing static transaction snapshots, the system transforms data into time-based sequences and applies difference transforms to capture temporal patterns. This dimensional transformation enables the detection of evolving merchant behaviors and MCC shifts that traditional static models miss, improving detection accuracy without proportionally increasing processing complexity.
2Reliability
If merchants shift transaction volumes across MCCs to improve authorization rates, then authorization benefits can be gained, but the reliability of the authorization approval process deteriorates due to gaming the system
Solution Approach 1:
The patent implements a feedback mechanism that continuously monitors merchant transaction patterns across MCCs and time periods. The system generates anomaly scores based on detected deviations from expected patterns and feeds this information back into the authorization process. This feedback loop enables the system to adapt to and detect merchant attempts to game the system by shifting MCCs, maintaining authorization process integrity while allowing legitimate merchant flexibility.
Solution Approach 2:
The patent applies preliminary action by training machine learning models on historical transaction data before actual detection occurs. The system pre-processes and transforms historical data, trains models to recognize normal versus anomalous patterns, and establishes baseline expectations for merchant behavior. This preliminary preparation enables the system to quickly and reliably detect MCC shifting attempts in real-time without compromising authorization process integrity.
3Measurement precision
If time series analysis and machine learning are applied to detect MCC shifts, then detection accuracy improves, but the device complexity increases due to additional processing steps
Solution Approach 1:
The patent segments the complex analysis task into distinct modular components: data collection, time series transformation, difference transform application, model training, and anomaly scoring. Each component handles a specific aspect of the analysis independently. This modular segmentation reduces system complexity by making each component manageable and replaceable, while collectively achieving high detection accuracy through their coordinated operation.
Solution Approach 2:
The patent replaces traditional mechanical rule-based detection systems with machine learning models that automatically learn patterns from data. Instead of manually configuring complex detection rules, the system uses trained models to identify anomalies. This substitution reduces operational complexity while maintaining or improving detection accuracy, as the models adapt to new patterns without requiring system reconfiguration.
Data Source
AI summary
Systems, methods, and computer program products for detecting merchant data shifts may identify a shift in transaction volume of a merchant system across Merchant Category Codes (MCCs) using a combination of time series analysis and machine learning.