Comparison the Performance of Classification Methods for Diagnosis of Heart Disease and Chronic Conditions

In this paper, we aim to identify the best ratio of training set, and to evaluate the diagnostic performance of eight classical machine learning methods for chronic diseases. The five categories of chronic disease datasets were collected from UCI and GitHub database include heart disease, breast cancer, diabetic retinopathy, Parkinson’s disease and diabetes. Machine learning (ML) methods including six individual learners (logistic regression (LR), BP neural network (BP), learning vector quantization (LVQ), extreme learning machine(ELM), support vector machine (SVM), decision trees (DTs)), and two ensemble learning methods were implemented using BP_Adaboost (Ada) and random forest (RF). We used five indicators, including AUC value, accuracy, sensitivity, specificity and running time to compare the behaviors of each algorithm. The ensemble learning methods were the most suited for the diagnosis of chronic diseases and led to a significant enhancement of performance compared to individual learners.