Heart Disease Risk Level Prediction: Knitting Machine Learning Classifiers

Contemporary medical diagnosis is a multifaceted process, requiring accurate patient data, clinical expertise acquired over several years and a philosophical perceptive of relevant medical literature. The numerous uncertainty risk factors which characterises heart disease mean that the diagnosis of this disorder is a complex task, even for the experts. In an effort to decrease both the time required for disease diagnosis as well as to enhance the accuracy of the diagnosis, clinical decision support systems (DSS) have been developed that incorporate data mining techniques to enhance the disease diagnosis accuracy. When literature is investigated, it has been observed that different researchers report the classifier's performance based on the overall accuracy of the classifier. That is, most researchers would evaluate the classifier's performance and choose the best classifier based on its overall accuracy. The question raised by this study is “What if the best classifier based on the overall accuracy is not a good predictor of a particular class in question within the dataset?” This paper thus presented the diagnosis of heart disease risk level using classification techniques under the data mining approach. A detailed comparative study focuses on how different classifiers predict each heart-disease class during the process of disease diagnosis. The comparative study is discussed using the confusion matrix of each classifier.