Predicting Autism Spectrum Disorder using Machine Learning Technique

Abstract

Autism Spectrum Disorder (ASD) is a psychiatric disorder that puts constraints on the ability to use of cognitive, linguistic, communicative, and social skills. Recently, many data mining techniques employed to serve this domain by determining the main features of the condition and the correlation between them. In this article, we investigate the Association Classification (AC) technique as a data mining technique in predicting whether an individual has autism or not. Accordingly, seven well-known algorithms are selected to conduct analysis and evaluation of the performance of the AC technique in term of identifying correlations between the features to help decide early on whether an individual has autism; this is particularly significant for children. The evaluation for the behavior and the performance in the prediction tasks for the AC algorithms was conducted for the common metrics of including Precision, Accuracy F-Measure as well as Recall. Finally, a comparative performance analysis among the algorithms was used as final result for the study. The results show better performance for the WCBA algorithm in most test scenarios with accuracy of 97 % although, the majority of algorithms exhibited excellent accuracy when applied in this domain.