Explainability techniques for Artificial Intelligence models in medical diagnostic

The integration of artificial intelligence (AI) techniques into clinical settings presents critical challenges due to the opacity of machine learning models, often referred to as "black boxes": this study explores the application of explainability techniques, specifically Local Interpretable Model-agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP), in the context of medical diagnostics. Using the "Diabetes Health Indicators Dataset", we applied Logistic Regression as a predictive model to identify key risk factors for diabetes and evaluate the ability of explainability techniques to improve transparency and interpretability. The results demonstrate that SHAP provides a detailed global and local understanding of feature importance, offering clinicians insights into key predictors such as HighBP, CholCheck, and GenHlth; LIME complements this by delivering intuitive explanations for individual predictions, enabling rapid and accessible interpretation. The combination of these techniques enhances trust in AI systems by providing both comprehensive insights and actionable explanations. Challenges related to computational complexity, scalability, and the integration of these methods into clinical workflows are also discussed, along with recommendations for future research aimed at developing scalable, interpretable AI models for ethical and responsible medical use.