Heart Disease Classification with XAI and Kernel SHAP

Heart disease is a major worldwide health concern, necessitating advances in precise and cost-effective diagnostics. While machine learning (ML) models have shown promise in this field, their opaque nature frequently impedes clinical use due to a lack of interpretability. This work investigates the effectiveness of Explainable AI (XAI) approaches, specifically Kernel SHAP, to increase the transparency and trustworthiness of machine learning models for heart disease categorization. We evaluated the interpretability and performance of a logistic regression model using a well-known heart disease dataset. Our logistic regression model has an accuracy of 85.2%, precision of 82.5%, recall of 84.7%, and F1-score of 83.6%. Kernel SHAP analysis gave useful insights, demonstrating that characteristics like ’ca’ (number of main arteries) and ’thal’ (thalassemia) have the high influence on predictions. These findings suggest that adding XAI approaches might significantly increase the confidence and acceptance of ML models in healthcare contexts, eventually leading to better patient outcomes.