Explainable AI and Federated Learning in Financial Fraud Detection

Financial fraud detection has become increasingly challenging due to the growth of digital transactions, the sophistication of fraudulent behavior, and strict requirements for data privacy and regulatory (ACM, 2024; AWS ML Blog, 2025) compliance. Traditional centralized machine learning approaches struggle with issues of data sharing restrictions, lack of transparency, and scalability across financial institutions. Recent advances in Federated Learning (FL) and Explainable AI (XAI) offer promising solutions to these challenges. FL enables collaborative fraud detection without exposing sensitive customer data, as demonstrated in studies utilizing meta-learning, graph-based federated architectures, and cross-framework implementations (Zheng et al., 2020; Tang et al., 2024; Abdul Salam et al., 2024). At the same time, XAI methods such as SHAP (Zhou et al., 2023; JISEM Journal, 2025), LIME, and user-centered design approaches improve interpretability, fostering trust among stakeholders and compliance with regulatory (ACM, 2024; AWS ML Blog, 2025) frameworks (Zhou et al., 2023; JISEM, 2025; Zhou, 2025). Integrating FL with XAI creates systems that are both privacy-preserving and transparent, as seen in hybrid approaches employing explainable ensemble models, secure federated frameworks, and real-world case studies on platforms like AWS SageMaker (Aljunaid et al., 2025; Almalki et al., 2025; AWS ML Blog, 2025). Surveys and reviews further highlight how these technologies address explainability (Almalki et al., 2025; Aljunaid et al., 2025), robustness, and fairness (Ojo et al., 2025; Zhou, 2025) while maintaining high detection accuracy (Zheng et al., 2020; Tang et al., 2024) (ACM, 2024; Wang et al., 2025). Collectively, this body of work indicates that explainable federated fraud detection frameworks can provide scalable, trustworthy, and regulation-aligned solutions for modern financial systems. Future research should focus on enhancing cross-institutional collaboration, integrating adversarial (Wang et al., 2025; Deshmukh et al., 2025) robustness, and tailoring explanations to diverse user needs in financial ecosystems.