Comparison of Tree Based Model Feature Importances With SHAP Scores

Heart failure, which affects over 64 million individuals globally, is a serious health problem. Early and accurate prediction is crucial for reducing its impact. Researchers have employed various machine learning models to address this challenge. Our aim is to find whether the importance scores generated by random forest, gradient boost, and XGBoost agree with the explain-ability method. This paper shows the prediction model whose feature importances agree with the Shapley values for our selected dataset. We also compare feature importance scores of three prediction algorithms random forest, gradient boost, and XGBoost with SHAP (Shapley Additive exPlanations) to find their agreement on scores. We use the UCI “Heart Failure Clinical Records” collection from the UCI Machine Learning Repository, which has $\mathbf{2 9 9}$ patients in its heart failure dataset.