Heart Disease Prediction using Ensemble Learning

Early disease prediction prevents patients to encounter severe situations. Transition from symptoms to disease stage of any disease takes time. Prediction of disease in this transition stage is called early detection. For the heart disease risk prediction in this study, an ensemble learning system is recommended. The stack ensemble learning approach is used to predict cardiac illness using a variety of heterogeneous weak learners. These include Multi-Layer Perceptron Classifier, Decision Tree Classifier, Support Vector Classifier, and Logistic Regression. All of these ML techniques are layered in the stack-based Ensemble classifier. Using stack-based ensemble classification, the dataset is utilized to distinguish between the presence and absence of early heart disease symptoms. The complete dataset is used as input for all weak learning ML algorithms. A meta-classifier is used in the ensemble learning technique known as stacking to combine many classification models. Performance evaluation for all ML algorithms is constructed using the Logistic Regression metaclassifier-based technique. SMOTE is used to handle unbalanced data, as determined by the classification of the data. The dataset needed for disease prediction is taken from UCI repository. A Python-based implementation of the proposed method is carried out. The ensemble approach approached 0.922 (AUC) in the ROC-AUC curve. The performance is evaluated using confusion matrix and classification reports.