Implementing Ethical AI Governance Framework with SHAP and LIME for Enhanced Cybersecurity in Financial Institutions

This research presented an explainable AI-based fraud detection framework using the XGBoost classifier, supplemented with SHAP and LIME for interpretability. The framework was created and evaluated using the Synthetic Financial Datasets for Fraud Detection. Experimental results demonstrated very high accuracy (99.97%) with strong precision (0.95+), recall (0.84+), and F1-score (0.89+) values across various configurations. The "black box" nature of XGBoost was resolved by incorporating SHAP and LIME into the AI pipeline. SHAP offered a global perspective of feature importance, while LIME provided local explanations by dissecting individual transaction predictions. This degree of explainability is essential for regulatory compliance in the financial sector, building stakeholder trust, error analysis and system auditing, and iterative model enhancements. Future work should include: testing on real-world anonymised financial datasets; exploring ensemble approaches with LightGBM and CatBoost; implementing online learning algorithms for concept drift; adding domain-specific features such as geolocation inconsistency and device fingerprinting; deep learning sequence models for temporal fraud patterns; real-time deployment on cloud platforms; and adversarial robustness testing. The ethical implications with respect to GDPR, FATF, and the EU AI Act compliance should also be investigated.

[1] Sabharwal, R., Miah, S.J., Wamba, S.F., et al. (2024) 'Extending application of explainable artificial intelligence for managers in financial organizations', Annals of Operations Research. https://doi.org/10.1007/s10479-024-05825-9. [2] Lee, S.U., Perera, H., Liu, Y., et al. (2024) 'Integrating ESG and AI: A comprehensive responsible AI assessment framework', arXiv:2408.00965v2. [3] Mohsin, M.T. and Nasim, N.B. (2025) 'Explaining the unexplainable: A systematic review of explainable AI in finance', arXiv:2503.05966. [4] Sayles, J. (2024) 'The current state of AI governance and model risk management', in Principles of AI Governance and Model Risk Management. Apress. https://doi.org/10.1007/979-8-8688-09. [5] Saleem, M., Chawla, C., et al. (2025) 'Responsible AI in Fintech: Addressing challenges and strategic solutions', in Generative AI in FinTech. Springer. https://doi.org/10.1007/978-3-031-76957-3_4. [6] Yang, W., Wei, Y., Wei, H. et al. (2023) 'Survey on Explainable AI: From Approaches, Limitations and Applications Aspects', Hum-Cent Intell Syst 3, 161-188. [7] Elmaghrabi, M. (2024) 'ESG risk management, sustainability-oriented firms and financial distress in the financial industry', 25th ACIT, pp. 1-3. [8] Mhammad, A.F., et al. (2023) 'Generative & responsible AI - LLMs use in differential governance', CSCI 2023, pp. 291-295. [9] Agbese, M., et al. (2021) 'Governance of ethical and trustworthy AI systems', IEEE REW 2021, pp. 224-229. [10] Findlay, M. & Seah, J. (2020) 'An ecosystem approach to ethical AI and data use', IEEE AI4G, pp. 192-197.