Comparison of Machine Learning Models for Automated Autism Diagnosis

Abstract

Autism Spectrum Disorder (ASD) is characterized by impairment in communication and language skills as well as repetitive and stereotyped behaviors. Early ASD diagnosis helps in developing a meaningful outcome in its treatment. Machine learning (ML) models can provide faster diagnostic capacity to determine patterns not observable by humans through behavioral analysis. We applied the ML classification models, including random forest, logistic regression, K-nearest neighbor intuition, support vector machine, decision tree, kernel support vector machine, and Naive Bayes, for each data set (children, teenagers, and adults). Our results show that ML models are powerful tools that can assist healthcare professionals in diagnosing ASD. Our model predicts non-autism cases with 97.9% accuracy. We believe that performing a logistic regression analysis indicating which factors increase or decrease the probability of diagnosis is a significant contribution. We hope to elucidate alternative ways to objectively diagnose ASD for timely treatment purposes.