Abstract:
Diagnosing Autism Spectrum Disorder (ASD) is difficult due to the unique symptoms displayed by each individual. This study examined nine commonly used Machine Learning (ML) classifiers, Random Forest (RF), Support Vector Machine (SVM) Linear, SVM with RBF kernel, AdaBoost classifier, k- Nearest Neighbour (kNN), Gradient Boosting (GB), Logistic Regression (LR), Gaussian Na¨ıve Bayes (GNB), and Multilayer Perceptron (MLP) classifier, across three ASD datasets for adults, adolescents, and children. The classifiers’ effectiveness was measured by evaluating their accuracy, F1 score, recall, precision, sensitivity, and specificity across various testing sizes (10%, 20%, and 30%). The findings suggest that ML has the potential to improve ASD screening, which could supplement traditional diagnostic methods. Among the classifiers, SVM Linear, AdaBoost, and LR consistently achieved remarkable performance, reaching 100% accuracy across all testing sizes and datasets. The results show the robustness of these classifiers in accurately identifying ASD patterns, making them strong candidates for ASD prediction models. The MLP classifier also showed competitive performance, indicating the importance of data split choice on the effectiveness of classifiers. These results underscore the potential of ML in enhancing ASD screening and diagnosis, suggesting a promising direction for further research in the application of these techniques on a broader and more diverse range of ASD data.