Enhancing predictive modelling and interpretability in heart failure prediction: a SHAP-based analysis

Predictive modelling plays a crucial role in healthcare, particularly in forecasting mortality due to heart failure. This study focuses on enhancing predictive modelling and interpretability in heart failure prediction through advanced boosting algorithms, ensemble methods, and SHapley Additive exPlanations (SHAP) analysis. Leveraging a dataset of patients diagnosed with cardiovascular diseases (CVD), we employed techniques such as synthetic minority over-sampling technique (SMOTE) and bootstrapping to address class imbalance. Our results demonstrated exceptional predictive performance, with the gradient boosting (GBoost) model achieving the highest accuracy of 91.39%. Ensemble techniques further enhanced performance, with the voting classifier (VC), stacking classifier (SC), and Blending achieving accuracies of 91.00%. SHAP analysis uncovered key features such as time, Serum_creatinine, and Ejection_fraction, significantly impacting mortality prediction. These findings highlight the importance of transparent and interpretable machine learning models in healthcare decision-making processes, facilitating informed interventions and personalized treatment strategies for heart failure patients.