The Evolution of AI-Driven Predictive Modelling in Fintech: A Systematic Survey of Risk, Fraud, and Cybersecurity

Abstract

As the financial services sector transitions to real-time, cross-border digital architectures, traditional risk management frameworks face unprecedented challenges from increasing transaction velocity and sophisticated, AI-driven threats. This paper provides a comprehensive systematic survey of the academic literature published between 2021 and 2026, focusing on the convergence of Artificial Intelligence (AI) in risk management, fraud detection, and cybersecurity.Through a PRISMA-guided review of over 30 core studies, we categorize the evolution of methodologies from static statistical models to dynamic, multimodal intelligence systems. Key findings indicate a significant shift toward the use of Graph Neural Networks (GNNs) for detecting systemic contagion and the integration of Large Language Models (LLMs) for proactive market sentiment analysis. Furthermore, this survey explores the critical "Accuracy-Explainability" trade-off, analysing how Explainable AI (XAI) and Federated Learning have emerged as essential tools for regulatory compliance under the EU AI Act and DORA frameworks.The analysis reveals that while AI significantly enhances predictive accuracy, it also introduces new vulnerabilities, specifically regarding adversarial machine learning and synthetic identity fraud. We conclude by identifying persistent research gaps, including the need for quantum-resistant security protocols and more robust defences against generative adversarial deception. This study serves as a strategic roadmap for researchers and practitioners navigating the "arms race" between defensive AI and industrialized financial crime.