USING OPTIMAL MACHINE LEARNING ALGORITHMS TO PREDICT HEART FAILURE PATIENT CLASSIFICATION

Abstract

Heart failure (HF) remains one of the leading causes of mortality worldwide, making early prediction and diagnosis essential for improving patient survival and reducing healthcare costs. Machine learning (ML) techniques have demonstrated considerable potential in assisting clinicians with accurate disease prediction. However, most heart failure datasets suffer from class imbalance, which negatively affects classification performance, particularly for minority class patients. This paper presents an optimized Extreme Gradient Boosting (XGBoost) model integrated with the Synthetic Minority Over-sampling Technique (SMOTE) for heart failure patient classification. Initially, missing values, outliers, and redundant attributes are removed through preprocessing. SMOTE is then applied to balance the dataset by generating synthetic minority samples. Hyperparameter optimization using Grid Search with Stratified Cross-Validation identifies the optimal XGBoost parameters. The proposed framework is evaluated using Accuracy, Precision, Recall, F1-score, ROC-AUC, and Matthews Correlation Coefficient (MCC). Experimental results demonstrate that the optimized XGBoost-SMOTE model significantly outperforms traditional machine learning algorithms including Logistic Regression, Decision Tree, Random Forest, Support Vector Machine, KNearest Neighbors, AdaBoost, and baseline XGBoost. The proposed approach achieves an accuracy of 98.21%, precision of 97.94%, recall of 98.47%, F1-score of 98.20%, and ROC-AUC of 99.10%, indicating superior predictive capability for heart failure diagnosis. These findings suggest that integrating SMOTE with optimized XGBoost provides an effective decision-support tool for clinical risk assessment. Similar findings have been reported in prior studies evaluating XGBoost with SMOTE-based preprocessing for heart failure prediction.