Abstract

The complex neurological condition known as autism spectrum disorder (ASD) is characterized by repetitive behaviors, social interaction, and communication. For immediate assistance and support, early and accurate ASD prediction is essential. In this study, we use a dataset of behavioral and clinical variables to assess how well different machine learning (ML) algorithms predict ASD. The algorithms analyzed include Decision Tree Classifier, Gaussian Naive Bayes (GNB), XGBoost, K-Nearest Neighbors (KNN), LightGBM, and CatBoost. Our findings show that sophisticated ensemble techniques perform more accurately than conventional classifiers. With an accuracy of 95.39%, the GNB classifier demonstrated a notable improvement over the Decision Tree (DT) classifier, which had an accuracy of 85.11%. The ensemble approaches XGBoost, LightGBM, and CatBoost, however, achieved the highest accuracies, with respective results of 97.87%, 97.16%, and 98.23%. With an accuracy of 93.26%, the KNN classifier likewise demonstrated strong performance. These findings suggest that ensemble methods, particularly CatBoost, provide superior predictive performance for ASD detection compared to other algorithms. The confusion matrix analysis further supports the robustness of these models by highlighting their precision and recall metrics. According to the study’s findings, applying advanced machine learning algorithms could significantly increase the predictive accuracy of ASD, perhaps resulting in an earlier diagnosis and better outcomes for those on the spectrum. Future studies should examine how these models might be incorporated into therapeutic settings and evaluate how applicable they are in the real world.

Authors

Sanju S. Anand¹, Shashidhar Kini²
Srinivas University, India¹, Srinivas Institute of Technology, India²

Keywords

Autism Spectrum Disorder, Decision Tree, Naive Bayes (NB), XGBoost, K-Nearest Neighbors, Machine Learning, LightGBM, CatBoost, Predictive Modelling, Ensemble Methods