Graph-Based Fraud Detection Using Neural Networks

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing fraud detection systems in transaction systems rely on static features, which fail to capture complex interdependencies between consumers and merchants, resulting in low accuracy in detecting fraudulent transactions.

Innovation Solution

A method and system that generate a consumer-merchant graph based on historical transaction data, determining both static and dynamic features, and using these features to train a neural network for real-time or near real-time fraudulent transaction detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If static features are used for fraud detection, then the system is simple to implement, but the detection accuracy is insufficient due to failure to capture complex interdependencies

Engineering Contradiction:
Improvefraud detection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent transitions from static, isolated transaction features to a graph-based representation that adds relational dimensions. By constructing a graph where nodes represent consumers and merchants and edges represent transactions, the system captures complex interdependencies and contextual relationships that static features cannot represent, thereby improving detection accuracy without excessive complexity increase

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Solution Approach 2:

The patent introduces graph embeddings as an intermediary that transforms complex graph structures into compact vector representations. These embeddings serve as a bridge between the complex relational data and the neural network classifier, enabling the system to capture sophisticated patterns while maintaining computational efficiency and manageable system complexity

Inventive Principle:
Principle #24Intermediary (Mediator)

2Reliability

If graph-based dynamic features are used, then complex interdependencies are captured, but computational complexity increases

Engineering Contradiction:
Improvefraud detection reliabilityVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent performs graph embedding computation in advance, transforming the complex graph structure into pre-computed vector representations. This preliminary action separates the computationally intensive graph processing from the real-time fraud detection task, allowing the system to capture complex interdependencies reliably while keeping the actual detection process computationally efficient

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent replaces direct computation with graph neural networks or complex algorithms with a more efficient approach using pre-computed graph embeddings. This substitution maintains the ability to capture complex relational patterns while significantly reducing the computational burden during fraud detection operations

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentEP3975092B1Method and system for detecting fraudulent transactions
Publication Date: 2025.04.30 MASTERCARD INT INC
  • EP3975092B1 patent drawingFigure 1A
  • EP3975092B1 patent drawingFigure 1B
  • EP3975092B1 patent drawingFigure 2A

AI summary

A method for detecting fraudulent transactions includes generation of a graph including a plurality of nodes and a plurality edges between the plurality of nodes based on historical transaction data of a plurality of historical transactions. The plurality of nodes include a set of merchant nodes and a set of consumer nodes. A set of static features and a plurality of dynamic features are determined based on the historical transaction data and the generated graph, respectively. A neural network is trained based on the set of static features and the plurality of dynamic features for detection of transaction fraud. The neural network is used to detect a first transaction as one of a fraudulent transaction or a legitimate transaction based on first transaction data of the first transaction.