Abstract:
Heart disease is a group of diseases that affect the heart and blood vessels. Someexamples are heart failure, arrhythmia's, valve disorders, and coronary arterydisease. It is a global health problem that needs to be found early and worked on together tobecontrolled and stopped. Heart disease is one of the major causes of mortality aroundtheworld, and hence early detection and severity determination are essential for propertreatment. The study aims at the application of supervised machine learning algorithmsfor the prediction of mortality risk and disease progression in patients with heart failureusing the Heart Failure Clinical Records dataset. Three machine learning algorithms—logistic regression, random forest, and XGBoost—were trained for classification(deathprediction) and regression (severity prediction). Cutting-edge feature engineeringtechniques, such as Principal Component Analysis (PCA), Shapely AdditiveExplanations (SHAP), and evolutionary algorithms, were employed in the selectionofsignificant predictors: age, serum creatinine, and ejection fraction. SHAPandLocal Interpretable Model-agnostic Explanations (LIME) were included to ensure model interpretability and clinical utility of the results. For classification tasks, theperformance was examined using precision, recall, accuracy, and ROC-AUC; forregression tasks, it was evaluated using mean squared error (MSE), root meansquarederror (RMSE), and R². With 85% accuracy, an ROC-AUC of 0.91 for mortalityprediction, and an R² of 0.75 for severity progression, XGBoost performed better thanthe other models. Logistic regression performed slightly worse compared torandomforest, which showed competitive performance. These results prove that XGBoost isauseful instrument for predicting the mortality and severity of heart disease whenpairedwith powerful feature engineering and interpretability techniques. Validatingthesemodels in larger, diverse cohorts and implementing them in medical settings shouldbethe main goals of future research.
