Machine Learning Algorithms for Early Prediction of Autism: A Multimodal Behavioral and Speech Analysis Approach

Posted: Dec 19, 2018

• Hammad Khan
  Punjab College
• Michael Johnson
  University of Illinois Urbana-Champaign
• Emily Smith
  University of Illinois Urbana-Champaign

Abstract

Autism Spectrum Disorder (ASD) represents a complex neurodevelopmental
condition characterized by challenges in social communication and restricted, repetitive behaviors. Early detection is crucial for timely intervention and improved
long-term outcomes, yet current diagnostic procedures often delay diagnosis until
after age four. This research presents a comprehensive machine learning framework for early autism prediction using multimodal behavioral and speech data.
We collected and analyzed data from 1,850 children aged 18-48 months, including
video-recorded behavioral interactions, vocal characteristics, and standardized developmental assessments. Our approach integrates feature extraction from multiple
modalities including prosodic features, speech fluency metrics, eye gaze patterns,
motor coordination, and social responsiveness indicators. We developed and compared multiple machine learning models including ensemble methods, deep neural
networks, and hybrid architectures. The proposed multimodal ensemble achieved
exceptional performance with 94.2% accuracy, 93.8% sensitivity, and 94.5% specificity, significantly outperforming single-modality approaches. Feature importance
analysis revealed that vocal prosody, joint attention behaviors, and speech disfluency patterns were the most discriminative predictors. This research establishes
an effective computational framework for early autism screening that could be deployed in clinical and educational settings to facilitate earlier identification and
intervention.