AI-Driven Fraud Detection in Financial Transactions with Graph Neural Networks and Anomaly Detection

Detecting fraudulent financial transactions is crucial for upholding the integrity of economic systems. Traditional methods often lag behind evolving fraud tactics, prompting the need for innovative approaches. We propose a pioneering framework combining Graph Neural Networks (GNNs) with anomaly detection techniques to enhance fraud detection. Transactions are represented as graphs, allowing GNNs to capture intricate fraud patterns. Anomaly detection methods flag suspicious transactions. Experimentation on the widely used Credit Card Fraud Detection dataset, comprising transactions made by European cardholders, showcases substantial advancements over baseline methods. The dataset is highly unbalanced, with fraudulent transactions accounting for only 0.172% of all transactions. Our approach achieves a detection rate of 95% with a false positive rate of 2%, surpassing the performance of the current state-of- the-art Gradient Boosting Classifier by 10%. It exhibits resilience against various fraud schemes, including account takeover and identity theft. Ablation studies underscore the significance of graphbased representations and anomaly detection mechanisms. Our research underscores the efficacy of GNNs and anomaly detection in bolstering financial fraud detection, presenting a promising solution against sophisticated fraudulent activities..

I. Introduction

Detecting fraudulent financial transactions is paramount for safeguarding economic systems and maintaining public trust in financial institutions. As the digital economy expands, so too do the methods employed by fraudsters, necessitating innovative approaches to combat financial crimes effectively. Traditional rule-based and statistical methods often struggle to keep pace with the evolving sophistication of fraudulent activities, prompting the exploration of advanced techniques such as Graph Neural Networks (GNNs) and anomaly detection.