An Optimized Machine Learning Framework for Heart Failure Patient Classification and Risk Prediction

Abstract: Heart failure is a major global health problem, and early prediction of patient risk can significantly improve treatment outcomes. This project presents a machine learning-based system for classifying heart failure patients into high-risk and low-risk survival categories using clinical data. The dataset consists of approximately 5000 patient records with important medical features such as age, ejection fraction, serum creatinine, and blood pressure. Data preprocessing techniques including feature scaling, feature selection using SelectKBest, and class balancing using SMOTE were applied to improve model performance. Multiple machine learning algorithms were evaluated, including Logistic Regression and XGBoost. Among them, the XGBoost model demonstrated the best performance, achieving an accuracy of 99.70% with high precision and recall. A Flask-based web application was also developed to allow users to input patient data and obtain real-time risk predictions.