Machine Learning for Enhanced Autism Screening: A Comparative Evaluation of Classification Algorithms

Detecting autism spectrum disorder (ASD) is challenging due to its varying symptoms across individuals. This research assessed four common machine learning classifiers - Logistic Regression, Naïve Bayes, Support Vector Machines, and K-Nearest Neighbors - using four ASD datasets. Two experiments evaluated the algorithms on the original datasets and versions modified by combining three datasets, adding an age feature, and reducing features. Support vector machines showed the most consistently strong performance on all metrics in both experiments. The adult dataset performed best among original datasets, while the modified combined dataset was most successful, with substantial gains across measures. Overall, the results demonstrate machine learning's promise for developing more accurate and accessible ASD screening, complementing standard diagnostic practices. Further research should evaluate these techniques on expanded and diverse ASD data.