Ensemble Learning Classification for Medical Diagnosis

According to a report by the World Health Organization(WHO) of diseases responsible for deaths, heart disease tops the chart. It is revealed that diagnosing heart disease in an earlier stage is an important issue and researchers across the world have investigated to develop intelligent support systems that could help a physician in better medical discernment beforehand. Machine Learning can be used to develop a medical diagnosis system for predicting heart disease which not only promises a more accurate diagnosis but also reduces the cost of diagnosis. In this paper, we are automating the task of predicting heart disease by using statistical methods/machine learning algorithms. We are exploiting some of the most important algorithms used these days like Logistic Regression, Support Vector Machine (SVM), Decision trees and also Ensemble Learning algorithms (Bagging, Boosting and Random Forests). We conduct these experiments on Cleveland Heart Dataset available on UCI KDD Archive. The dataset distribution judged by our research fits well with decision trees and gives better results with ensemble learning methods, we justify this claim by comparing ensemble results with other Machine Learning approaches. Random forest classifier works best for this data set distribution, the random-nature of this ensemble approach fits the dataset well providing good test accuracies of upto 96.26%.