Adversarial Resilient And Explainable Ai Framework For Deepfake-Aware Fraud Detection In Financial Systems

Abstract

This study presents an Adversarial Resilient and Explainable AI Framework for Deepfake-Aware Fraud Detection in Financial Systems, addressing the dual challenges of adversarial robustness and interpretability. The methodology integrates adversarial trained deep learning modules, explainable AI mechanisms, and fault detection resilience layers, enabling accurate identification of transaction-level anomalies and deepfake-induced manipulations. Experimental evaluation demonstrates substantial performance gains, with accuracy improving from 0.82 to 0.95, precision from 0.80 to 0.93, recall from 0.78 to 0.92, F1-score from 0.79 to 0.93, and AUC-ROC from 0.83 to 0.96. Comparative analysis with baseline systems highlights the superiority of the proposed framework in mitigating false positives and false negatives under adversarial stress conditions. The inclusion of explainability ensures transparency in decision-making, reinforcing regulatory compliance and stakeholder trust. Overall, the framework establishes a scalable and secure solution for combating deepfake-enabled financial fraud in real-world deployment scenarios.