Graph-Based Fraud Detection via Network Relationship Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing fraud detection systems fail to detect fraudulent activity in newer payment methods, such as real-time payments, and often mislabel legitimate activity as fraudulent due to their focus on individual accounts rather than network interactions.
Innovation Solution
The implementation of a graph-based fraud detection system that generates a graph data structure from transactions between accounts, determines features of the graph, and updates account profiles based on new transactions, allowing for more accurate detection of fraudulent behavior by analyzing semantic relationships within the payment network.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing fraud detection systems focus on individual accounts, then the system complexity is low, but the fraud detection accuracy deteriorates
Solution Approach 1:
The patent transitions from analyzing accounts in isolation (one-dimensional) to analyzing accounts within a network context (multi-dimensional). By constructing a graph where accounts are nodes and transactions are edges, the system adds the dimension of relational connectivity, enabling detection of fraudulent patterns that span multiple accounts and would be invisible in individual account analysis.
Solution Approach 2:
The patent merges individual account data with network-level transaction data to create a comprehensive fraud detection model. By combining node-level features (account attributes) with graph-level features (network patterns, community structures), the system achieves more accurate fraud detection than either approach could provide alone.
2Measurement precision
If graph-based analysis is implemented to improve fraud detection accuracy, then the measurement precision improves, but the computational complexity increases
Solution Approach 1:
The patent segments the fraud detection process into distinct phases: graph construction from transaction data, feature extraction at multiple levels (node, edge, community, graph), and model training. This segmentation allows each phase to be optimized independently and enables parallel processing of feature computation, reducing overall computational burden.
Solution Approach 2:
The patent performs preliminary actions by pre-computing graph features and storing them in a structured format before fraud detection is needed. Graph profiles are built and updated incrementally as new transactions arrive, so that when fraud detection is required, the computational work has already been done, enabling faster real-time analysis.
3Productivity
If real-time payment fraud detection is implemented, then the productivity improves, but the measurement precision deteriorates due to limited data availability
Solution Approach 1:
The patent performs preliminary analysis by constructing graph profiles and computing features from historical transaction data before real-time fraud detection is needed. This pre-computation creates a baseline understanding of normal account behavior and network relationships, enabling accurate real-time detection even with limited new data available.
Solution Approach 2:
The system incorporates feedback mechanisms where detection results and updated transaction data continuously refine the graph profiles. As more transactions are processed, the graph structure and features are updated, improving the accuracy of real-time detection over time while maintaining the ability to provide immediate fraud assessment.
Data Source
AI summary
A method, system, and computer program product is provided for graph-based fraud detection. The system includes at least one processor programmed or configured to generate a graph data structure based on a plurality of transactions between a plurality of accounts, wherein each account of the plurality of accounts is represented by a node in the graph data structure, and wherein each transaction of the plurality of transactions is represented by an edge in the graph data structure, determine a plurality of features of the graph data structure for each account of the plurality of accounts, generate a graph profile for at least one account of the plurality of accounts based on the plurality of features for the at least one account, and update the graph profile for the at least one account based on at least one new transaction engaged in by the at least one account.