Fraud Detection Using Locality Sensitive Hashing and Local Outlier Factor
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for detecting fraudulent transactions are inadequate as they rely on static rules that fail to identify anomalies exhibiting different patterns, making it difficult to detect evolving forms of fraudulent data points in electronic transactions.
Innovation Solution
The use of a combination of Locality Sensitive Hashing and Local Outlier Factor algorithms to construct feature vectors, assign best neighbor data points, compute outlier scores, and detect fraudulent data points based on these scores, allowing for the identification of inconsistent patterns.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If static rule-based detection methods are used, then the system can detect known anomaly types, but it fails to detect anomalies with different or evolving patterns
Solution Approach 1:
The patent transitions from static rule-based detection to dynamic pattern recognition by continuously analyzing transaction data and adapting to new fraud patterns. The system uses evolving algorithms that learn from historical data and adjust detection criteria in real-time, enabling it to detect both known and emerging anomaly types without requiring manual rule updates.
Solution Approach 2:
The system changes detection parameters dynamically by adjusting sensitivity thresholds, time windows, and analysis dimensions based on the specific transaction context and detected patterns. This allows the system to optimize detection accuracy for different fraud types while maintaining adaptability to new patterns through parameter reconfiguration rather than fixed rules.
2Measurement precision
If comprehensive analysis of all transaction data is performed, then detection accuracy improves, but processing time and computational complexity increase
Solution Approach 1:
The patent segments the large transaction dataset into smaller, manageable clusters based on similarity metrics and transaction characteristics. By dividing the data into segments that can be analyzed independently and in parallel, the system achieves comprehensive analysis accuracy while reducing overall processing time through distributed computation and focused attention on relevant subsets.
Solution Approach 2:
The system performs partial analysis by initially examining a subset of transactions or features that are most indicative of fraud, then deepening the analysis only for suspicious cases. This selective approach achieves sufficient detection precision for high-priority transactions while avoiding the computational overhead of exhaustive analysis across all data points.
Data Source
AI summary
Methods and systems for detecting fraudulent data points in a database of a computerized system include receiving, from a user device, a request for detecting one or more fraudulent data points in a database storing data points representing electronic transactions. The system constructs a feature vector fore each data point and k random vectors. After constructions, the system performs a combination of Locality Sensitive Hashing algorithm and Local Outlier Factor (LOF) algorithm on the constructed feature vectors and the constructed k random vectors to compute outlier scores for each data point. The system detects a fraudulent data point based on the outlier scores that indicate consistent degrees.