COMPARISON OF MACHINE LEARNING MODELS IN HEART FAILURE PREDICTION AND THEIR INTEGRATION INTO CLINICAL DECISION SUPPORT SYSTEMS

Heart failure remains a leading cause of morbidity and mortality worldwide, necessitating advanced tools for early risk prediction. This study presents an interactive, machine learning-driven web application designed to predict heart failure outcomes using clinical data. Leveraging the heart failure clinical records dataset (n=299), the application integrates a comprehensive suite of fifteen diverse predictive models, encompassing traditional/statistical-based algorithms, instance-based and probabilistic methods, various tree-based and ensemble techniques, and neural networks within an intuitive Shiny framework. Key features include exploratory data analysis (correlation matrices, feature importance), model training, and real-time risk prediction with customizable patient parameters. The system employs stratified cross-validation (10-fold) for robust evaluation and achieves impressive performance, with top-performing models exhibiting test set Area Under Curve values exceeding 0.85, alongside high scores in accuracy, sensitivity, specificity, and F1-score. By combining clinical variables such as ejection fraction, serum creatinine, and follow-up time, the tool demonstrates how interactive machine learning platforms can enhance clinical decision-making. The open-source R-Shiny implementation provides immediate visual feedback, model interpretability features, and a template for extending predictive analytics to other medical domains. This work bridges the gap between statistical modeling and clinical application, offering both a prognostic tool and an educational resource for data-driven cardiology.