Hyperparameter Optimization of Machine Learning Models Using Differential Evolution for Heart Failure Mortality Prediction

Purpose: Heart failure is a cardiovascular disease associated with high morbidity and mortality rates. Various machine learning models have been increasingly applied to predict mortality among patients with heart failure, but their predictive performance can be influenced by hyperparameter configuration. This study aims to optimize hyperparameters using the Differential Evolution (DE) algorithm to improve predictive performance. Methods: This research used the Heart Failure Clinical Records Dataset, which consists of 299 patient data with 13 clinical features. The dataset is first split into a training set and a testing set, with the testing set kept independent throughout the modeling process. Within the training process, feature selection is performed using Chi-square-based SelectKBest, class imbalance is addressed using the Synthetic Minority Oversampling Technique (SMOTE), and feature scaling is performed using StandardScaler. These preprocessing steps are incorporated into the cross-validation pipeline to prevent information leakage. Next, nine machine learning algorithms were optimized using Differential Evolution with an accuracy based objective function and 10-fold cross-validation. Result: Experimental results show that Differential Evolution produced model-specific effects, with accuracy improvements observed in six of the nine evaluated machine learning models compared with their baseline configurations. Random Forest (RF), Support Vector Machine (SVM), and Extra Trees Classifier (ETC) achieved the highest accuracy of 0.8500, while RF obtained the highest F1-score and ROC-AUC of 0.7692 and 0.8922, respectively. The exact McNemar test showed that none of the baseline-versus-DE differences reached statistical significance at the 0.05 level. Novelty: This research proposes a mortality classification framework for heart failure patients that integrates SelectKBest, SMOTE, and Differential Evolution for hyperparameter optimization across nine machine learning models. The framework enables a systematic evaluation of the model specific effects of Differential Evolution on heart failure mortality classification performance.