Advancing Heart Failure Prediction: A Comparative Study of Traditional Machine Learning, Neural Networks, and Stacking Generative AI Models

Heart failure (HF) poses critical global health challenges, emphasizing the need for robust predictive models to support early diagnosis and enhance patient outcomes. Traditional machine learning (ML) models, such as Logistic Regression (LR), Support Vector Machines (SVM), Random Forests (RF), Gradient Boosting Machines (GBM), and Extreme Gradient Boosting Machines (xGBM), have shown effectiveness but face limitations in handling nonlinear relationships, addressing class imbalances, and generalizing across datasets. Deep learning (DL) models, including Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs), excel at identifying complex patterns but are hindered by computational requirements and limited interpretability, restricting clinical adoption. This research evaluates predictive models using nine datasets ranging from 299 to 400,000 records. Synthetic Minority Over-sampling Technique (SMOTE) was applied to address class imbalances, while a Stacking Generative AI (GenAI) model was developed. This hybrid model integrates Generative AI with RF, GBM, and CNNs, enhancing underrepresented subgroup representation through synthetic data generation. The Stacking Generative AI model demonstrated superior performance, achieving 98% accuracy and a Receiver Operating Characteristic Area Under the Curve (ROC AUC) of 0.999 on a 1,025-record dataset. These results highlight the model's ability to handle complex data, enhance predictive accuracy, and improve clinical relevance. A web application further illustrates its practical value, offering an accessible platform for HF risk assessment. This study underscores the innovative role of hybrid models in advancing healthcare decision-making and improving patient care.