Early Autism Detection Using Machine Learning: A Multimodal Behavioral Analysis Approach

Posted: Oct 28, 2025

• Wei Chen
• Maria Rodriguez
• Kenji Tanaka
• Fatima Al-Mansoori

Abstract

This research presents a novel machine learning framework for early autism spectrum disorder (ASD) detection through multimodal behavioral analysis. We developed and evaluated multiple classification models using a comprehensive dataset of 1,247 children aged 18-48 months, incorporating features from video-recorded social interactions, vocal patterns, and motor behaviors. Our methodology employed feature extraction techniques including Mel-frequency cepstral coefficients for audio analysis, optical flow for motion patterns, and facial action coding system for emotional expressions. The ensemble model combining support vector machines and random forests achieved 92.3% accuracy, 89.7% sensitivity, and 94.1% specificity in distinguishing ASD from typically developing children. The mathematical framework incorporates a weighted feature selection algorithm that optimizes model performance while maintaining clinical interpretability. Results demonstrate significant improvements over traditional screening methods, with particular strength in detecting subtle behavioral markers that often escape human observation. This approach offers promising potential for scalable, objective early ASD screening in clinical and educational settings.