Can machine learning models accurately predict the likelihood of Autism Spectrum Disorder (ASD) using questionnaire-based screening data and basic demographic features

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition characterized by difficulties in social interaction, communication, and repetitive behaviors. Early detection plays a crucial role in improving developmental outcomes through timely intervention. However, traditional diagnostic processes rely heavily on expert clinical evaluation and behavioral observation, which can be time-consuming and inaccessible in many regions. This project investigates whether machine learning models can assist in early ASD screening using questionnaire-based behavioral indicators and demographic features. Using a publicly available dataset from the UCI Machine Learning Repository, multiple supervised learning models, including Logistic Regression, Random Forest, XGBoost, and AdaBoost, are evaluated. The study emphasizes recall-oriented evaluation using metrics such as Recall, F1-score, F2-score, ROC-AUC, and PR-AUC to reflect the importance of minimizing false negatives in medical screening tasks. Exploratory data analysis and interpretability analysis are conducted to understand the relationship between behavioral responses and ASD classification. Results show that questionnaire-based behavioral features provide strong predictive signals and that recall-focused threshold optimization improves screening sensitivity. The study demonstrates that machine learning can serve as a decision-support tool for ASD risk screening while maintaining ethical and interpretability considerations.