International Journal of Scientific Research and Engineering Development

Financial fraud is rapidly increasing in volume and complexity as digital payments, mobile wallets, open banking, and real-time transactions expand worldwide. Traditional rule-based systems are often insufficient to detect adaptive and emerging fraudulent activities. Artificial Intelligence (AI) and Machine Learning (ML) have become essential tools, leveraging supervised learning for known fraud detection, unsupervised anomaly detection for novel patterns, and graph-based models to capture complex relationships among accounts, transactions, and devices.
This paper surveys recent advances (2020–2025) and proposes an integrated AI-driven framework combining: (1) feature-based ensemble classifiers for robust predictions, (2) graph neural networks (GNNs) to identify relational and collusive fraud, (3) federated learning for cross-institutional model training while preserving privacy, and (4) explainable AI (XAI) for interpretability and regulatory compliance. Experimental results indicate that graph-based and federated approaches outperform traditional featureengineered methods, offering higher precision, recall, and adaptability. Hybrid AI frameworks integrating ensemble learning, graph modeling, federated learning, and XAI provide adaptive, privacy-preserving, and transparent solutions for modern financial fraud detection.