Explainable Machine Learning Approach for Heart Failure Patient Survival Events Prediction

This research addresses the challenge of predicting the survival status of heart failure patients, a critical concern due to the global prevalence of heart failure. The primary objective is to construct an interpretable machine learning model utilizing data from Felege Hiwot referral hospital and Injibara general hospital in Amhara regional state, Ethiopia. The study employs diverse machine learning algorithms, such as Decision Trees (DT), Logistic Regression (LR), K-Nearest Neighbors (KNN), Deep Neural Networks (DNN), and XGBoost, and incorporates techniques like Grid Search Hyperparameter Optimization (HPO) and Synthetic Minority Over-sampling Technique (SMOTE) to enhance model performance. The findings demonstrate that XGBoost surpasses other methods, achieving an impressive 0.93 Area under the Curve (AUC) value. In addition, the SHAP and LIME helped to gain insight on the algorithms’ decision. This research holds significant implications for informed healthcare decision-making, given the substantial global burden of heart failure.