Fraud Detection Using Time-Ordered Event Patterns
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting fraudulent transactions at Point of Sale (POS) sites are inefficient due to high false alarm rates and inaccurate detection, particularly in retail environments where employee theft is prevalent, leading to significant retail shrinkage and growth hampering.
Innovation Solution
A system and method utilizing a computation engine that generates time-ordered discrete text labels from video surveillance and transaction log data, parses these labels to identify true and false patterns, and employs pattern classification techniques like Neural Networks and Markov Models to categorize transactions as fraudulent or genuine in real-time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If frequency-based voting mechanism is used to classify transactions, then detection coverage is improved, but false alarm rate increases
Solution Approach 1:
The patent transforms the classification approach by changing from frequency-based voting to pattern-based classification using multiple parameters including temporal sequences, event patterns, and contextual relationships. This parameter transformation enables more accurate fraud detection while reducing false alarms by considering the temporal and contextual dimensions of transaction events rather than simple frequency counts.
Solution Approach 2:
The patent introduces temporal and contextual dimensions to transaction analysis by sequencing events chronologically and analyzing their relationships. This dimensional expansion from simple frequency-based classification to multi-dimensional pattern recognition improves detection accuracy while reducing false positives through more comprehensive event evaluation.
2Reliability
If video surveillance data is captured and stored for analysis, then detection capability is improved, but data processing time increases
Solution Approach 1:
The patent applies preliminary action by pre-processing video surveillance data during transactions to extract and sequence relevant events in real-time. This preliminary extraction and temporal sequencing of key events enables rapid pattern recognition and classification without requiring extensive post-analysis of raw video data, thus improving detection capability while minimizing analysis time.
Solution Approach 2:
The patent extracts only the essential event information from video surveillance data through automated event detection and temporal sequencing. This extraction of critical features rather than analyzing complete video streams reduces processing time while maintaining detection capability by focusing on discriminative event patterns.
3Measurement precision
If pattern classification techniques are applied to transaction data, then detection accuracy is improved, but system complexity increases
Solution Approach 1:
The patent segments the complex pattern classification task into distinct components: event detection, temporal sequencing, pattern extraction, and classification. This segmentation of the analysis pipeline into manageable stages reduces implementation complexity while maintaining high classification accuracy by addressing each aspect with specialized processing.
Data Source
AI summary
A system and a method for detecting fraudulent transactions at a transaction site by analyzing pattern of events associated with one or more transactions are provided. The present invention provides for forming a collection of most probable fraudulent patterns and true patterns associated with one or more transactions, selecting a pattern classification technique, generating a data input from an ongoing transaction that is interpretable by the selected pattern classification technique, and effectively and efficiently categorising ongoing transaction into fraudulent and genuine transactions using selected pattern classification technique. The present invention may be utilized in a variety of applications where discrete time-ordered visual events are associated with a transaction, for example: vehicles detected in relation to a transit point, badge or card swipes from an automated door lock etc., which indicate trespassing, theft and unauthorized access to restricted areas etc. with a primary focus on retail shrinkage.