Optimizing Diabetic Coronary Heart Disease Prediction Models Using Ensemble Learning Approaches

Abstract

Diabetic Coronary Heart Disease (DCHD) continues to be a global threat, taking millions of lives every year. The potential of healthcare data on heart disease to influence decisions is still largely unrealized, despite the data being available in massive quantities. The identification of cardiovascular conditions, such as heart attacks and coronary artery disorders, presents a formidable obstacle that traditional clinical data analysis finds difficult to overcome. This work aims to improve the accuracy of weak classification algorithms and demonstrate the usefulness of the method in early diabetic heart disease prediction by implementing it on a medical dataset. In response, this research introduces a novel heterogeneous ensemble learning approaches (ELA), to forecast cardiac disease early by utilizing a novel combination comprising four base classifiers adaptive boosting (AdaBoost), K-Nearest Neighbour (KNN), Random Forest (RF), Support Vector Machine (SVM) to improve the forecast results' accuracy. The two-feature selection wrapping approaches in this article are backward elimination and forward selection. Proposed method is compared with artificial neural networks (ANN), support vector machines (SVM), decision tree (DT), Naive bayes (NB) and eXtreme Gradient Boosting (XGBoost). Evaluation metrics, such as F1 score, recall, accuracy, precision, are employed to assess the proposed method. The experimental results surpassed other methods with classification accuracies of 91% and 92% on the Cleveland and Framingham datasets, respectively. Additionally, the ROC curve analysis showed that the ensemble method had higher ROC (Receiver Operating Characteristic) values, indicating better performance compared to other methods. The results demonstrated that the challenge of classifying an unbalanced diabetic heart disease dataset might be solved using an ensemble-learning framework based on other models.