FRAUD DETECTION IN FINANCIAL SYSTEMS USING EXPLAINABLE AI

Abstract

This article presents a systematic review of AI driven fraud detection methodologies, evaluating the efficacy and transparency limitations of supervised, unsupervised and deep learning paradigms. Central to this study is the examination of Explainable Artificial Intelligence (XAI) as a reconciliation layer, specifically analyzing feature attribution methods (SHAP, LIME) and counterfactual explanations. Drawing on empirical evidence from industrial case studies, including Danske Bank and JPMorgan Chase, the research validates that interpretability significantly mitigates false positive rates, but faces challenges regarding computational latency and cognitive load. Synthesizing these findings, the paper proposes a novel Human-in-the-Loop (HITL) architectural framework that integrates XAI tools directly into analyst workflows. This approach advocates for a synergistic socio-technical system, where machine scalability is augmented by human ethical judgment, ensuring a fraud detection ecosystem that is robust, legally compliant and operationally transparent.