AUTDETECT: A Machine Learning-Based Web Tool for the Early Detection of Autism Spectrum Disorder in Children

Abstract

Early diagnosis of ASD is crucial for timely intervention, especially in low-resource environments where access to specialized evaluation is limited. This study aimed to develop and validate a web-based application, supported by machine learning algorithms, to assist in the early detection of ASD using behavioral screening questionnaires. A responsive web application was designed using Django (backend) and React (frontend), deployed on Amazon Web Services. The system collects responses to the Q-CHAT-10 questionnaire from caregivers and uses multiple supervised learning models to predict ASD risk. The data used for model training and evaluation were obtained from a public ASD screening dataset. Data preprocessing, SMOTE for class balancing, and hyperparameter tuning through GridSearchCV were applied. Clinical validation was performed through pilot testing at a hospital in Lima, Peru. Among the tested models, the Support Vector Machine, Random Forest, and XGBoost classifiers achieved the highest performance, with F1-scores exceeding 0.90. The system showed a 80% reduction in processing time for the clinical evaluation process compared to the traditional workflow. Clinicians reported improved efficiency and usability, and the application demonstrated strong potential for scalable deployment in similar clinical settings. The proposed web-based system is a valuable tool for supporting early ASD detection in under-resourced environments, as its combination of validated screening tools and machine learning predictions enhances diagnostic workflows, enabling earlier intervention and better clinical decision-making.