Accuracy improvement for diabetes disease classification: a case on a public medical dataset

Abstract

As a chronic disease, diabetes mellitus has emerged as a worldwide epidemic. Providing diagnostic aid for diabetes disease by using a set of data that contains only medical information obtained without advanced medical equipment, can help numbers of people who want to discover the disease or the risk of disease at an early stage. This can possibly make a huge positive impact on a lot of peoples lives. The aim of this study is to classify diabetes disease by developing an intelligence system using machine learning techniques. Our method is developed through clustering, noise removal and classification approaches. Accordingly, we use SOM, PCA and NN for clustering, noise removal and classification tasks, respectively. Experimental results on Pima Indian Diabetes dataset show that proposed method remarkably improves the accuracy of prediction in relation to methods developed in the previous studies. The hybrid intelligent system can assist medical practitioners in the healthcare practice as a decision support system.