Performance Comparison of SVM Kernel Types on Child Autism Disease Database

Abstract

In this work, a database was used under the name Autistic Spectrum Disorder Screening Data for Children Data Set which was acquired from the data warehouse (UCI database repository). This dataset contains information for 292 children with 21 attributes. Using Weka tool. Mentioned data were classified by whether is diagnosed with autism disease or not. Using four types of support vector machine kernels. Normalized polynomial kernel, polynomial kernel, PUK kernel and RBF kernel classifiers utilized in data mining. The values which were used for performance comparisons are accuracy, precision, sensitivity, F measure and confusion matrix for each kernel. In this study, 100% successful results of accuracy have been obtained with each of polynomial kernel and PUK kernel.