Klasifikasi Pengidap Diabetes Pada Perempuan Menggunakan Penggabungan Metode Support Vector Machine dan K-Nearest Neighbour

Abstract

Diabetes Mellitus is a metabolic disease with characteristics of hyperglycemia that occurs due to abnormalities in insulin secretion, insulin action or both. The detection of diabetes mellitus disease using the dataset Pima Indians had been done by various methods, one of which implementation methods is K-Neaarest Neighbor (KNN). One drawback of the KNN method is the determination of the optimal parameters k. Value of k that are too high will reduce the effect of noise on the classification, but makes the boundaries between each classification is becoming increasingly blurred, while the value of k that is too low will result in sample taking values for the less and lead to reduced accuracy. For this study proposes the use of Support Vector Machine (SVM) as the optimal solution of k determination. In this study, we will implement the hybrid SVM-KNN method to be used as a method of classification of people with diabetes using the dataset "Pima indian". Experiments done by varying the parameter values and the kernel used to see the value of the accuracy of the hybrid SVM-KNN method. Parameters that influence the value of C, tolerance, sigma, bias and the value of k on KNN. The highest average value of the accuracy obtained by using SVM-KNN is 92.00% and proved to be better than traditional SVM method average of the accuracy only 77.60% and KNN is 91%.