Application of Graph Machine Learning Techniques for Transaction Fraud Detection

Summary

Fraud detection in financial transactions constitutes a critical issue in the field of finance, as the rapid advancement of technology has increased both the frequency and complexity of fraudulent activities. Traditional data analysis methods often prove inadequate in addressing the growing intricacy of these problems. In this context, graph machine learning algorithms are emerging as an innovative and powerful approach to fraud detection, as they enable the exploitation of structural features and relationships within data.This thesis examines the application of graph machine learning algorithms in the detection of fraud in financial transactions.

It analyzes modern techniques such as Graph Neural Networks (GNNs) and Graph Convolutional Networks (GCNs), while also investigating their advantages and limitations in real-world scenarios. By incorporating features such as graph topology, relationship analysis, and data dynamics, the thesis presents a framework aimed at improving the accuracy and efficiency of fraudulent transaction detection. The research findings suggest that graph machine learning algorithms can reveal hidden fraud patterns, offering a robust tool for combating fraud in financial transactions. The study concludes with recommendations for future research and potential applications in industrial settings.