Transforming Early Autism Screening with Multimodal Machine Learning in Digital Health

Early and accurate detection of Autism Spectrum Disorder (ASD) is the key factor in facilitating early interventions and enhancing development in children with ASD. This paper proposes an innovative, multimodal diagnostic framework that utilizes machine learning methods by combining both behavioral testing through questionnaires and facial image analysis. Each modality was trained independently, and the results were fused in a user-friendly mobile application, increasing accessibility as well as clinical utility. The proposed system demonstrates high performance in identifying ASD-related characteristics, achieving high predictive accuracy through a non-invasive, cost-effective, and scalable method. Our results demonstrate the transformative potential of machine learning to improve early ASD screening (especially in resource-limited contexts) and contribute to the development of intelligent digital health solutions. The integrated model was found to have an overall diagnostic accuracy of 85%, with the questionnaire-based classifier having almost perfect accuracy and the image-based model showing a promising generalizability. These findings show that the proposed system has the potential to become a practical solution for early ASD detection and effectively enable timely, life-changing interventions.