XAI Framework Enhances Banking Anomaly Detection for Auditors.

The banking sector increasingly relies on automated systems to detect fraudulent transactions, but traditional rule-based methods often suffer from high false-positive rates and lack transparency, making them difficult for compliance and audit teams to use effectively. This limitation hinders trust and efficient investigation. Researchers have developed an Explainable Artificial Intelligence (XAI) framework specifically designed for anomaly detection in banking transactions, tailored for internal audit workflows. The framework employs an Isolation Forest (iForest) model to identify anomalous transactions, which is then augmented by a SHAP (SHapley Additive exPlanations) layer. This SHAP layer provides clear, transaction-level explanations, attributing the anomaly score to specific features based on cooperative game theory. A user-friendly Streamlit dashboard presents these explanations in an accessible format for audit professionals, eliminating the need for machine learning expertise. Evaluation on a synthetic banking dataset showed high precision (0.91) and recall (0.88), outperforming several unsupervised baselines. Expert feedback confirmed that these feature-level explanations significantly boost auditor confidence and improve decision-making quality, paving the way for more accountable and transparent AI deployment in regulated financial environments.