Comparison of Classification Algorithms of the Autism Spectrum Disorder Diagnosis

Abstract

ASD sufferers face difficulties in early development compared to normal humans. Various tools, clinical, and non-clinical approaches have been implemented but take a long time to produce a complete diagnosis. the solution by adopting machine learning. This study proposes the application of cross-validation techniques in the Decision Tree method, Linear Discriminant Analysis, Logistic Regression, SVM, and KNN and determines the best k value in each classification method because the shift of datasets when using cross-validation techniques in the classification method is one factor that can cause the estimate to be inaccurate. The results show that the decision tree provides an accuracy of 100% in each of the k values that have been determined previously. 96.9% on Linear Discriminant Analysis with $k=7, k=9$, and $k =10$. 99.7% in Logistic Regression with values of $k=2$ and $k= 3$. 99.9% in Support Vector Machine with values of $k=9$ and $k =1\theta$ and 94.2% for K-Nearest Neighbors with a value of $k=8$.