Siamese Neural Network for User Activity Data Breach Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Machine learning approaches face challenges in accurately and timely categorizing user account activity data, particularly in distinguishing authorized from unauthorized transactions, due to user behavior drift and the need for tolerance in deviation detection to avoid false alerts.
Innovation Solution
A trained classifier uses machine learning techniques to analyze user activity data by selectively removing sub-groups of data to identify significant differences, employing a Siamese neural network with dropout methods to determine account breaches, and comparing recent transactions with historical data to isolate fraudulent activity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If machine learning models are trained to closely identify differences in user data to detect unauthorized transactions, then detection precision improves, but false alerts increase due to user behavior drift
Solution Approach 1:
The patent segments user activity data into multiple sub-groups and selectively removes certain sub-groups during classification. This allows the system to identify which specific data segments contribute most to classification differences, thereby improving detection precision while accounting for normal behavior variations that would otherwise cause false alerts.
Solution Approach 2:
The patent applies different classification thresholds and analysis methods to different sub-groups of user data. By treating different segments of user behavior differently based on their characteristics, the system can maintain high detection precision for suspicious patterns while being more tolerant of normal behavioral drift, thus reducing false alerts.
2Measurement precision
If the system raises alerts for any deviation in user behavior to ensure security, then detection sensitivity improves, but system complexity increases due to managing multiple classification thresholds
Solution Approach 1:
The patent enables the classification system to automatically determine which sub-groups of data to remove and what thresholds to apply, based on analysis of the data itself. This self-adjusting mechanism allows the system to maintain high detection sensitivity without requiring manual configuration of multiple complex thresholds, thereby reducing system complexity.
3Productivity
If traditional classification methods are used to categorize user activity data, then processing speed is maintained, but detection accuracy decreases for subtle fraudulent patterns
Solution Approach 1:
The patent performs preliminary segmentation and analysis of user data into sub-groups before the main classification process. By pre-organizing the data and identifying key segments in advance, the system can maintain processing speed while improving detection accuracy for subtle fraudulent patterns that would be difficult to detect in raw, unsegmented data.
Data Source
AI summary
Machine learning techniques can be used to train a classifier, in some embodiments, to accurately detect similarities between different records of user activity for a same user. When more recent data is received, newer data can be analyzed by selectively removing particular sub-groups of data to see if there is any particular data that accounts for a large difference (e.g. when run through a classifier that has been trained to produce similar results for known activity data from a same user). If a sub-group of data is identified as being significantly different from other user data, this may indicate an account breach. Advanced machine learning techniques described herein may be applicable to a variety of different environments.