A predictive approach for myocardial infarction risk assessment using machine learning and big clinical data

Abstract

Myocardial infarction is one of the most common cardiovascular diseases in emergency departments. Early prevention of this dangerous condition significantly impacts public health and considerable socioeconomic outcomes. The emergence of electronic health records (EHR) and the availability of real-world clinical data have provided opportunities to improve the quality and efficiency of healthcare by using artificial intelligence tools. In this study, we focus on the early recognition of risk factors, which can provide valuable information for early prediction of myocardial infarction and promoting a healthy life. Based on a big clinical dataset, we develop a predictive analytics approach for myocardial infarction. A vital step in efficient prediction is assessing the significance of input features, their relationships and their contributions to the disease. Therefore, we adopted statistical techniques, principal component analysis (PCA) and feature engineering. To reveal patterns and insights on our dataset, we implemented machine learning (ML) models varying from classical to more sophisticated: decision trees (DT), random forests (RF), gradient boosting algorithms (GBoost, LightGBM, CatBoost, and XGBoost) and deep neural networks (DNN). The imbalance-data issue is tackled by employing random under-sampling technique. The light gradient boosting model (LightGBM) with feature engineering on the balanced dataset is the best prediction performance achieved in this study.