Anomalous Transaction Detection via Multi-Window ML Transformations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional techniques for detecting money laundering activity are limited by their reliance on short duration moving windows, leading to low discriminative salience and high false positive rates, and can be easily circumvented.
Innovation Solution
A system and method for detecting anomalous activity related to potential fraud or criminal activity among a plurality of transactions, utilizing a machine learning model to apply various transformations such as value weighted netting, maximum frequency weighted netting, equal frequency weighted netting, and funneling weighted netting to identify potentially anomalous activity across linked bank accounts.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If conventional techniques use short duration moving windows to detect money laundering, then the detection speed is improved, but the discriminative salience deteriorates and false positive rates increase
Solution Approach 1:
The patent divides the transaction monitoring into multiple time windows of different durations (short, medium, long windows) that operate in parallel. Each window captures different temporal patterns of money laundering activity, allowing the system to maintain high detection speed through short windows while improving discriminative salience through longer windows that capture more comprehensive patterns.
Solution Approach 2:
The patent adds the temporal dimension by analyzing transactions across multiple time scales simultaneously. Instead of using a single time window, the system creates a multi-dimensional temporal analysis structure where short, medium, and long windows provide different perspectives on transaction patterns, enhancing both speed and precision.
2Measurement precision
If conventional techniques lower the detection threshold to improve sensitivity, then more anomalies are detected, but the number of false positives increases
Solution Approach 1:
The patent segments the detection process into multiple stages corresponding to different time windows. Each window applies its own thresholding logic, and the final detection decision is made by aggregating results across all windows. This segmentation allows the system to be sensitive to various patterns without triggering false positives from any single window.
Solution Approach 2:
The system incorporates feedback mechanisms where detection results from different time windows inform each other. The aggregation process uses feedback from multiple windows to adjust the overall detection decision, allowing the system to maintain high sensitivity while filtering out false positives through cross-validation across different temporal scales.
3Ease of operation
If conventional techniques use simple netting rules to improve ease of operation, then the system complexity is reduced, but the ability to detect sophisticated money laundering patterns deteriorates
Solution Approach 1:
The patent segments the complex detection task into simpler sub-tasks handled by different time windows. Each window applies relatively simple netting rules, but the combination of results from multiple windows achieves high detection accuracy for sophisticated patterns without requiring any single window to be overly complex.
Solution Approach 2:
The patent merges the results from multiple simple analysis windows to achieve complex detection capabilities. By combining the outputs of short, medium, and long windows, the system achieves high detection accuracy for sophisticated money laundering patterns while maintaining operational simplicity at each individual window level.
Data Source
AI summary
The present disclosure provides various transformations to be used in analysis of a large number of transactions to detect anomalies that would indicate potential fraudulent or criminal activity. Such transformations may be applied, for example, using a machine learning system. According to some examples, each of various transformations may be used to detect a particular type of behavioral anomaly. When multiple disparate transformations are considered together by the machine learning system, anomalous activity related to potential fraudulent or criminal activity can be detected more frequently and with greater accuracy.