Predicting Mortality from Heart Attack Clinical Data Using a Mixture of Experts Approach

Heart failure remains a leading cause of mortality worldwide, necessitating the development of accurate predictive models for early risk assessment. This study explores the application of a Mixture of Experts (stacking) approach by integrating Logistic Regression and Random Forest Classifier models to predict mortality among heart attack patients. The dataset, obtained from Kaggle's heart failure clinical records, comprises 299 patient records with 13 clinical features, including age, ejection fraction, serum creatinine, and high blood pressure. The models were trained and evaluated using an 80-20 data split, with performance measured using AUC, accuracy, precision, recall, and F1-score. While Logistic Regression achieved an accuracy of 81.67% and an AUC of 0.8240, the Random Forest model obtained a slightly higher AUC of 0.8274 but a lower accuracy of 75.00%. The ensemble meta-model improved discrimination power, achieving the highest AUC of 0.8446, demonstrating the effectiveness of combining multiple predictive models. Feature importance analysis revealed that time (follow-up period), serum creatinine, and ejection fraction were the most influential factors in predicting mortality. The results suggest that ensemble learning provides a more robust approach to clinical risk prediction, offering enhanced predictive accuracy and reliability over individual classifiers.