A Machine Learning Framework for Age-Inclusive Autism Spectrum Disorder Detection Using U.S. Screening Data

The growing prevalence of Autism Spectrum Disorder (ASD) is thus prompting the development of efficient, accurate, and accessible diagnostic tools for all ages. Although these include traditional diagnostic methods, such as the Autism Diagnostic Observation Schedule (ADOS) and Autism Diagnostic Interview-Revised (ADI-R), these methods are resource-intensive and appear to have been designed primarily for early childhood, leaving adolescents and adults (<tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{4 0 - 5 5}$</tex> years) at risk of missed diagnosis. We aim to develop a machine learning (ML) based framework to boost ASD detection in the U.S. Autism Screening Dataset across disparate age groups. The dataset comprises 704 records, including demographic, behavioral, and diagnostic attributes. Some data preprocessing techniques, normalized and encoded, improved inconsistency and the model performance. Then, principal component analysis (PCA) and correlation analysis were applied to determine the most relevant diagnostic features. Four supervised ML models over the Decision Tree, Random Forest (Gini and Entropy criteria), and Extra Trees Classifier are trained and evaluated using accuracy, precision, recall, f1 score, and ROC-AUC as performance metrics. Among these, the Extra Trees Classifier surpassed the other models, with 96.67 % accuracy, 97.50 % precision, 95.00 % recall, and 96.17 % F1 score, proving to be a good model to predict and robust. This highlights the potential of ML-driven approaches to complement traditional diagnostic frameworks by providing scalable, efficient, and age-independent detection of ASD. This framework addresses existing gaps by improving diagnostic accuracy and accessibility, which will lead to earlier intervention and better outcomes for people of all ages with ASD.