Coronary Heart Disease Prediction: A Novel Deep Learning-based Approach

Abstract

Coronary Heart Disease (CHD) is reported to be one of the world’s deadliest diseases. Early prediction or detection and diagnosis of the disease can help prevent, cure, and reduce the damage it could cause. Artificial intelligent techniques such as Machine Learning and Deep Learning have proven useful for the early detection and prediction of disease. However, issues of irrelevant and redundant features in open datasets have contributed to low classification accuracy rates and high misclassification rates. This leaves a gap for continuous approaches for smart feature selection and high-performing models in the field for better results. This study investigates the impacts of cardiologist-inspired datasets and PCA dimensionality reduction techniques on the performance of CHD prediction. The compared results show that while PCA improves the CHD prediction accuracy for datasets obtained from cardiologists, there exist no statistically significant differences in the efficacy of the PCA method for CHD classification when applied to open datasets, however, MLP and LSTM offer promising results. The results further indicated that the effectiveness of expert-based feature selection techniques on CHD classification is relatively stable when compared with open-source datasets.