Ethical AI Redefined: Pioneering Transparency with Hybrid Explainable Models
J. Gayathri, K. S. Vimal Chandran, F. Divya Rathna Mala, M S Aadyaanjali, Somkuan Kaviya, A. Maheshwaran · 2025
The rapid proliferation of artificial intelligence (AI) in critical domains like healthcare, finance, and governance has necessitated the development of ethical, transparent, and responsible AI systems. This study proposes a hybrid explainable AI model to achieve ethical, transparent, and high-performance artificial intelligence solutions. Combining black-box models such as Transformers with explainable techniques like SHAP, the proposed methodology addresses key challenges in AI, including fairness, interpretability, and privacy. The model tested on a diverse and representative dataset, focusing on performance, bias detection, and explainability. Results demonstrate that the proposed model achieved a 97.71% accuracy with an F1-score of 0.96, outperforming baseline models such as CNN (87.50%) and Transformer-Causal combinations (94.10%). SHAP analysis identified key features like Age, Blood Pressure, and Cholesterol Level as the most influential predictors, ensuring transparency and trustworthiness. Fairness evaluation across sensitive attributes, including gender and ethnicity, showed Demographic Parity of 96.40% and Equal Opportunity of 94.40%, indicating minimal bias. Privacy-preserving techniques such as Differential Privacy and Homomorphic Encryption implemented, achieving robust data protection with minor trade-offs. This hybrid approach successfully balances performance, fairness, interpretability, and privacy, presenting a scalable framework for ethical AI development.