Death Prediction of Heart Failure Patients Using Machine Learning Techniques

Abstract

Heart failure is a complex and debilitating condition that affects millions of people worldwide. Accurate prediction of heart failure progression is crucial for timely interventions and improved patient outcomes. This study compares the effectiveness of two predictive approaches: a statistical model and a domain knowledge-based model. Both models were trained using three machine learning algorithms-XGBoost, Logistic Regression, and Support Vector Machine-and evaluated using standard performance metrics. Our findings underscore the importance of incorporating both statistical data and domain-specific knowledge in heart failure diagnosis, as both models achieved comparable results. The statistical model achieved a Receiver Operating Characteristic Area Under the Curve (ROC AUC) score of up to 86.33% and a cross-validation score of up to 93.78%, while the domain knowledge model attained a ROC AUC of up to 83% and a cross-validation score of up to 93%. These comparable performance outcomes suggest that both types of data contribute meaningfully to forecasting heart failure progression. Furthermore, the study demonstrates that supervised machine learning techniques can accurately predict survival outcomes in heart failure patients using only a select set of patient attributes.