Ensemble Machine Learning Models for Optimized Clinical and Laboratory-Based Prediction of Heart Failure

Abstract

Heart failure remains a leading cause of both hospitalization and mortality in all countries of the world, and it is important to note the need to develop predictive models that would help physicians to detect and evaluate outcomes early. Kaggle Heart Failure Clinical Records was used, which includes demographic, clinical, and laboratory data. The Chi-square statistics was performed to select features and tests were performed on 13 and 7 feature subsets respectively. Class imbalance was corrected using Synthetic Minority Oversampling Technique (SMOTE). A wide range of classifiers, including Random Forest, Support Vector machine, AdaBoost, Gradient Boosting, and Stochastic Gradient Descent has been used with hyperparameters optimized using Artificial Immune Whale-Particle Swarm Optimization (AIW-PSO), which ensures high performance and efficient optimization of hyperparameters. Random Over Sampling (ROS) was explored to complement SMOTE and ensemble Voting Classifiers using LGBM and AdaBoost were used on the 13-feature and 7-feature sets. eXplainable AI methods, such as Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive explanations (SHAP), enhanced model interpretability, whereas deployment preparedness was demonstrated with the help of the Flask framework. According to experiments, the ROS-based 13F Voting Classifier had an accuracy of 99.5, and the 7F Voting Classifier had an accuracy of 99.0, demonstrating the effectiveness of the best ensemble learning in cardiac failure prediction.