Diagnosing Autism Spectrum Disorder Using Machine Learning Techniques

Abstract

Autism is a generalized pervasive developmental disorder that can be characterized by language and communication disorders. Screening tests are often used to diagnose such a disorder; however, they are usually time-consuming and costly tests. In recent years, machine learning methods have been frequently utilized for this purpose due to their performance and efficiency. This paper employs the most eight prominent machine learning algorithms and presents an empirical evaluation of their performances in diagnosing autism disorder on four different benchmark datasets, which are up-to-date and originate from the QCHAT, AQ-10-child, and AQ-10-adult screening tests. In doing so, we also utilize precision, sensitivity, specificity, and classification accuracy metrics to scrutinize their performances. According to the experimental results, the best outcomes are obtained with C-SVC, a classifier based on a support vector machine. More importantly, in terms of C-SVC performance metrics even lead to 100% in all datasets. Multivariate logistic regression has been taken second place. On the other hand, the lowest results are obtained with the C4.5 algorithm, a decision tree-based algorithm.