Explainable AI in action: a comparative analysis of hypertension risk factors using SHAP and LIME

Hypertension, a common and complex cardiovascular disease associated with a high risk of mortality and morbidity, has been targeted for detection using Artificial Intelligence Methods in recent times. However, due to their black-box nature, the reasons behind hypertension could not be identified by doctors. Therefore, there is an urgent need to elucidate the connections between hypertension and other biomarkers. In this study, local interpretable model-agnostic explanations (LIME) and SHapley Additive exPlanations (SHAP) were employed to demystify the hypertension risk prediction made by an extreme gradient boost (XGBoost) model. A comprehensive case record of 623 patients, encompassing reported hypertension and other diseases, medication usage, and laboratory results for 13 critical biomarkers, was analyzed in our study. The exemplary performance of our XGBoost model, demonstrating an accuracy of 99.4%, precision of 100%, recall of 97.30%, and an F1-score of 98.6%, further confirms the potential of Machine Learning (ML) in healthcare. Furthermore, the Biogeography-Based Optimization (BBO) algorithm was utilized to identify an effective subset of features. Almost half of the initial features were selected by the BBO algorithm. Achieving an accuracy, precision, recall, and F1-score of 97.7%, 96.9%, 93.9%, and 95.4%, respectively, using only 14 features selected by BBO highlights its efficacy. Additionally, our study demonstrated the impact of features on ML models through LIME and SHAP analyses. The preliminary system developed in this study underscores advanced, interpretable predictive modeling for early hypertension detection, laying the groundwork for enhanced risk assessment in diagnosis and preventive medicine in the future.