A Deep Learning–Based Predictive Model for Myocardial Infarction Risk Forecasting Using Patient Comorbidities and Lifestyle Data

Myocardial infarction is one of the most common causes of morbidity and mortality across the globe, complicated by intricate associations between physiological parameters, comorbidities and lifestyle habits. In the proposed study, a large-scale cardiovascular dataset of 281,934 patient records having 35 clinical features was used by the deep learning–based models for risk prediction. Random forest (RF), Naive Bayes, TabNet and one-class support vector machine were evaluated. Incorporating nonlinear feature transformation by the exponential linear unit (ELU) prior to RF classification (ELU on RF), with better accuracy, categorizes the normal and high-risk patients. Investigations were carried out to evaluate whether the root cause was an imbalanced dataset. The synthetic minority oversampling technique worked on the severe imbalance of the dataset, and Shapley additive explanations and LIME were helpful in finding a target column. Results suggested early prediction of the risk and bringing evidence-based preventive care.