AN OPTIMIZED XGBOOST-BASED FRAMEWORK FOR HEART FAILURE RISK CLASSIFICATION

Abstract

Heart failure continues to be one of the world's top causes of death, requiring reliable predictive models to enable prompt medical interventions. In order to solve class imbalance, this study offers a machine learning framework for heart failure survival prediction that makes use of an optimized XGBoost model combined with the Synthetic Minority Over-sampling Technique (SMOTE). To find the most significant predictors, we used Select K Best with Chi-square for feature selection using a dataset of 5000 clinical records that included features including age, ejection fraction, and serum creatinine. After analyzing several algorithms (such as Logistic Regression, Decision Tree, KNN, SVM, and Random Forest), the XGBoost model was chosen. It was then optimized to maximize hyperparameters, resulting in a test accuracy of 99.70%, precision, recall, and F1-scores close to 1.00, and an AUC-ROC of 0.9998. Our method performs better than the baseline Gradient Boosting Machine (GBM) with Adaptive Inertia Weight Particle Swarm Optimization (AIW-PSO) from earlier research, which attained 94% accuracy on a smaller dataset (299 patients). This is probably because of the larger dataset and sophisticated preprocessing. In addition to providing a scalable, high-accuracy tool for heart failure prognosis, this study demonstrates the effectiveness of XGBoost in conjunction with SMOTE for clinical predictive tasks and has the potential to enhance patient outcomes through accurate and prompt clinical decision-making.