A Comprehensive Survey on AI-Driven Financial Fraud Detection: Trends, Techniques, and Future Directions
TL;DR
This review provides a structured framework for developing intelligent, secure, and trustworthy fraud detection systems in financial institutions and serves as a roadmap for researchers and practitioners aiming to address evolving fraud threats in cloud-based, distributed, and privacy-sensitive environments.
Abstract
Financial fraud detection remains a critical challenge in digital ecosystems. This survey presents a comprehensive review of artificial intelligence (AI)-driven fraud detection techniques from 2015 to 2025, analyzing over 60 key studies. The evolution is traced across three phases: Early stage (classical machine learning (ML)), rule-based systems, data imbalance), algorithmic growth (deep learning (DL)), temporal analysis, hybrid models), and smart innovation (graph neural networks (GNNs)), federated learning (FL), explainable AI (XAI)). Major challenges include severe data imbalance, dynamic fraud patterns, black-box interpretability, privacy preservation, and adversarial robustness. Emerging techniques such as GNNs, FL, and reinforcement learning (RL) have shown superior performance in detecting organized and real-time fraud. Six future directions are proposed: Multimodal data fusion, continual learning (CL), large language models (LLMs) for textual anomaly detection, blockchain integration, human-in-the-loop explainable systems, and standardized evaluation benchmarks. This review provides a structured framework for developing intelligent, secure, and trustworthy fraud detection systems in financial institutions. It serves as a roadmap for researchers and practitioners aiming to address evolving fraud threats in cloud-based, distributed, and privacy-sensitive environments. Received: 14 November 2025 | Revised: 9 February 2026 | Accepted: 7 June 2026 Conflicts of Interest The authors declare that they have no conflicts of interest to this work. Data Availability Statement The data that support the findings of this study are openly available in the UCI Machine Learning Repository at https://archive.ics.uci.edu/. Author Contribution Statement Elham Shamsinejad: Formal analysis, Investigation, Data curation, Writing - original draft, Writing - review & editing. Hamid Banirostam: Conceptualization, Methodology, Validation, Writing - review & editing, Supervision, Project administration.