Abstract The increasing adoption of machine learning models in high-stakes domains such as healthcare has raised significant concerns regarding model transparency and interpretability. While complex models often achieve high predictive performance, their decision-making processes remain difficult to understand. Explainable Artificial Intelligence (XAI) techniques such as SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations) have been developed to address this challenge by providing post-hoc explanations of model predictions. This study presents a comparative evaluation of SHAP and LIME using two healthcare datasets—the Breast Cancer Wisconsin dataset and the Heart Disease UCI dataset—across three machine learning models: Logistic Regression, Random Forest, and Neural Network. The methods are evaluated using quantitative metrics including consistency, stability under input perturbation, and Spearman rank correlation against an interpretable baseline. The experimental results demonstrate that SHAP consistently outperforms LIME in terms of stability and consistency across multiple datasets and model architectures, suggesting that SHAP provides more reliable explanations for high-stakes healthcare applications. Keywords : Explainable AI, Machine Learning, SHAP, LIME, Model Interpretability, Healthcare AI, High-Stakes Systems
Prachi Puthran (Sun,) studied this question.