Fraud Detection System Using Graph-Based Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current monitoring systems are only able to detect about half of fraudulent cases in email money transfers, necessitating a more effective method for real-time detection.
Innovation Solution
A system and method utilizing a trained model to access real-time transaction data, extract graph-based and statistical features, and determine an account proximity score, combining machine learning with graph technology to identify fraudulent transactions within milliseconds.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current monitoring systems are used for fraud detection, then the system complexity is low, but the detection accuracy is only about 50%
Solution Approach 1:
The system segments the fraud detection process into multiple independent components: graph feature extraction module, statistical feature extraction module, model training module, and real-time scoring module. Each module handles a specific aspect of analysis, allowing the complex detection task to be divided into manageable segments that can be processed independently and in parallel.
Solution Approach 2:
The system transitions from traditional single-dimensional transaction analysis to multi-dimensional analysis by incorporating graph-based features (account relationships, transaction networks) alongside statistical features. This adds new dimensions to the detection space, enabling the system to identify fraud patterns that are invisible in conventional analysis.
2Measurement precision
If real-time transaction data is enriched with historical data and graph features, then the detection accuracy increases by 26%, but the processing time increases
Solution Approach 1:
The system performs feature extraction and data enrichment in advance during off-peak periods. Graph features and statistical features are pre-computed and stored for rapid retrieval during real-time transaction processing. This preliminary preparation eliminates the need for time-consuming computations during critical real-time detection moments.
Solution Approach 2:
The system replaces traditional rule-based mechanical detection methods with machine learning models that automatically learn and apply detection patterns. The trained model processes enriched features through intelligent algorithms, substituting complex mechanical rule evaluation with adaptive computational reasoning that handles rich feature sets more efficiently.
3Productivity
If the system processes 300 requests per second with comprehensive feature extraction, then the detection coverage improves, but the computational power required increases
Solution Approach 1:
The computational workload is segmented across multiple processing units and distributed systems. Feature extraction, model inference, and scoring operations are divided into separate processing stages that can be executed in parallel across different computational nodes, reducing the computational burden on any single unit while maintaining high throughput.
Solution Approach 2:
The system dynamically adjusts processing parameters based on transaction risk profiles and system load. For low-risk transactions, simplified feature sets are used with faster processing. For high-risk transactions, the full comprehensive feature extraction is applied. This parameter adaptation allows the system to maintain high throughput while optimizing computational resource utilization.
Data Source
AI summary
A computer system for, and method of, detecting fraudulent electronic transactions is provided. The system comprises at least one processor and a memory storing instructions which when executed by the processor configure the processor to perform the method. The method comprises accessing a trained model, receiving real-time transaction data, extracting graph-based and statistical features to enrich the real-time transaction data, and determining an account proximity score for the real-time transaction data.


