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.
E. Shamsinejad, Hamid Banirostam· FinTech and Sustainable Inno...· 0 citations
Digital transformation has sharply increased fraud risks in banking, with global annual losses exceeding $50 billion. This systematic review analyzes 20 influential peer-reviewed studies from 2024 to 2025, selected through a Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020–guided search from over 150 papers in IEEE Xplore Digital Library, SpringerLink, ScienceDirect, Scopus, and Google Scholar, using terms such as "AI banking fraud detection," "deep learning financial fraud," "graph neural network fraud," and "explainable AI banking." Included studies are published in reputable venues, report robust evaluations on real or large-scale transaction data, introduce meaningful innovations (e.g., novel architectures or privacy-preserving methods), and address deployment in banking environments. The reviewed works employ deep neural networks, graph-based models, hybrid ensembles, explainable artificial intelligence (XAI), and blockchain-inspired components, achieving detection accuracies of 93–96% on real transaction datasets. Nonetheless, major obstacles still limit large-scale adoption. We identify 10 recurring challenges; high computational costs, data quality problems, and limited model interpretability each appear in at least 30% of the studies. Our analysis compares the strengths, weaknesses, and contributions of existing approaches and highlights critical gaps in scalability, robustness, and standardization. We recommend wider use of lightweight models, broader application of federated learning, creation of shared benchmark datasets, and adoption of common evaluation protocols. Finally, we outline priority directions to move beyond a narrow focus on accuracy toward fraud detection solutions that are scalable, secure, transparent, and cost-effective, enabling more confident deployment of AI-based systems in real-world banking.
Received: 14 November 2025 | Revised: 14 February 2026 | Accepted: 14 July 2026
Conflicts of Interest
The authors declare that they have no conflicts of interest to this work.
Data Availability Statement
Data sharing is not applicable to this article as no new data were created or analyzed in this study.
Author Contribution Statement
Hamid Banirostam: Conceptualization, Methodology, Formal analysis, Investigation, Writing – original draft, Writing – review & editing, Visualization, Supervision, Project administration. Elham Shamsinejad: Conceptualization, Methodology, Validation, Formal analysis, Investigation, Data curation, Writing – original draft, Writing – review & editing.
Hamid Banirostam, E. Shamsinejad· FinTech and Sustainable Inno...· 0 citations
