Optimization of Accuracy to Autism Spectrum Disorder Identification for Children Using Support Vector Machine and Correlation-based Feature Selection

Abstract

Autism is a developmental disorder that affects the function of the neurological system, which can have a negative impact on the sufferer’s quality of life. The ratio of people with autism is relatively high and tends to increase, based on WHO in 2013 the ratio of Indonesian children suffering from autism is 1:160 or more than 112.000. Data mining is the one of data processing techniques that works to find patterns and knowledge from big data. One of the data mining techniques is a classification that works to search for models that reveal and estimate previously unidentified classes. SVM is one of the classification algorithms that use data to find the optimal hyperplane. SVM has the advantage to work well on a dataset that cannot be linearly separated. The disadvantage is that it can be challenging to select parameters that are ideal and to determine which ones have an impact and which ones do not. To reduce attribute dimensions, CFS was provided as a feature selection to improve accuracy based on correlation values. The Autistic Spectrum Disorder Screening for Children Dataset from the UCI machine learning repository was used in this research to compare the accuracy of SVM and CFS. The result of this research is the SVM algorithm yields an accuracy rate of 94.91%. When the SVM algorithm is combined with CFS, the accuracy rate rises to 96.61%, representing an improvement in accuracy of 1.7% by using the 17 selected attributes.