Diagnosis of diabetes mellitus using (chi square-information gain) selectors and (SVM and KNN) Classifiers

Abstract

In all age groups, one of the major diseases between individuals is Diabetes Mellitus (DM). Health-care industries are heavily depending on data mining in the diagnostics of diseases. Also, the medical data mining methods were utilized for the purpose of finding hidden patterns in datasets of medical domains in terms of medical treatment and diagnosis. The presented work is distinguishing normal or diabetic individuals with the use of 2 main phases. With regard to the 1st phase, feature selection was achieved with the use of information gain and chi-square test methods for finding the major efficient attributes regarding such disease. In terms of the 2nd phase, classification was conducted with the use of K-Nearest Neighbors (KNN) and Support Vector Machines (SVMs) algorithms. Pima India Dataset is used, in which it comprises (768) records, (268) positive predicted classes indicating diabetic patients and a total of (500) negative predicted classes indicate non diabetes. The experiment shows that SVM with Chi-square test give accuracy of 88% with the time taken in the implementation process was 0.02 seconds, KNN with Chi-square test give accuracy of 84% with the time taken in the implementation process was 0.03 seconds, and SVM with Information gain give accuracy of 87% with the time taken in the implementation process was 0.02 seconds, KNN with Information gain give accuracy of 82% with the time taken in the implementation process was 0.02 seconds.