Time-Dependent Graph Convolutional Network for Fraud Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional fraud detection systems are inaccurate, inflexible, and inefficient in determining whether digital identities originate from the same user, as they fail to consider attributes and are not inductive, requiring significant computing resources for retraining when changes occur.
Innovation Solution
A fraudulent transaction detection system utilizing a time-dependent graph convolutional neural network that generates node embeddings based on identity attributes and temporal dependencies, allowing for accurate and flexible identification of fraudulent transactions by analyzing interactions between digital identities in a heterogeneous network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional fraud detection systems are used, then computing resources are consumed for retraining when changes occur, but accuracy in determining whether digital identities originate from the same user deteriorates
Solution Approach 1:
The patent transforms the fraud detection approach by changing from traditional retraining-based systems to a graph neural network that processes identity relationships as graphical data. This parameter change enables the system to maintain high accuracy while reducing computing resource requirements, as the GNN naturally adapts to new identities through graph structure updates rather than full model retraining
Solution Approach 2:
The patent replaces the mechanical retraining process with a graph-based neural network that inherently handles identity relationships. The GNN substitutes the traditional machine learning retraining mechanism with a graph processing approach that automatically captures temporal dependencies and identity attributes, eliminating the need for frequent retraining while maintaining detection accuracy
2Adaptability or versatility
If conventional fraud detection systems are used, then system simplicity is maintained, but flexibility in adapting to changes deteriorates
Solution Approach 1:
The graph neural network serves multiple functions simultaneously: it detects fraud, captures temporal dependencies, processes heterogeneous identity attributes, and adapts to new identities all within a single unified framework. This multi-functionality provides the required flexibility without proportionally increasing system complexity
Solution Approach 2:
The patent implements a dynamic graph structure that evolves as new digital identities are encountered. The graph naturally adapts its structure and relationships over time, allowing the system to be flexible and responsive to changes in the fraud landscape without requiring manual reconfiguration or retraining
3Productivity
If conventional fraud detection systems are used, then training time is reduced, but productivity in identifying fraudulent transactions deteriorates
Solution Approach 1:
The graph neural network performs preliminary encoding of identity relationships and temporal patterns during graph construction. This preliminary action prepares the data in a format that enables rapid fraud detection without requiring subsequent retraining, thus improving productivity while minimizing time loss
Data Source
AI summary
The present disclosure relates to utilizing a graph convolutional neural network to generate similarity probabilities between pairs of digital identities associated with digital transactions based on time dependencies for use in identifying fraudulent transactions. For example, the disclosed systems can generate a transaction graph that includes nodes corresponding to digital identities. The disclosed systems can utilize a time-dependent graph convolutional neural network to generate node embeddings for the nodes based on the edge connections of the transaction graph. Further, the disclosed systems can utilize the node embeddings to determine whether a digital identity is associated with a fraudulent transaction.