| Summary: | Heart disease is one the leading causes of death globally, making the early detection of it crucial. Emerging technologies such as machine learning and deep learning are now being actively used in biomedical care, healthcare, and disease prediction. The focus of this paper is on the prediction of coronary heart disease (CHD) using a risk factor approach. Predictive techniques such as K-Nearest Neighbors, Binary Logistic Classification, and Naive Bayes are evaluated on the basis of metrics such as accuracy, recall, and ROC curves. These base classifiers are compared against ensemble modelling techniques such as bagging, boosting, and, stacking. A comparitive analytical approach was used to determine how ensemble techniques can be used to improve prediction accuracy of coronary heart disease. The modelling technqiues are tested on the ‘Cardiovascular Disease Dataset,’ which contains 70,000 records of patient data for coronary heart disease. Bagged models are shown to have an averaged increased accuracy of 1.96% in comparison to their traditional counterparts. Boosted models had an average accuracy of 73.4% but had the highest AUC score of 0.73. The stacked model involving KNN, random forest classifier, and SVM proved to be the most effective with a final accuracy of 75.1%. In addition, the perfomance of the tested models was validated using data-analytic technqiues and K-Folds cross-validation.
|
|---|