Autism Spectrum Disorder Diagnosis through Machine Learning: Techniques, Applications, and Clinical Challenges

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by impairments in social interaction, communication, and repetitive behavioural patterns. The growing demand for ASD in the world has intensified the requirements of early, accurate and scalable diagnostic methods. The standard modes of diagnosis are primarily based on behavioural assessments conducted by experts, which are time-consuming, subjective and often available in settings with low resources. The last several years saw the emergence of promising tools using Machine Learning (ML) and Artificial Intelligence (AI) to support and enhance ASD screening, diagnosis, and analysis with the use of data-driven processes. The review systematically examines previous research in terms of learning paradigms such as supervised learning, unsupervised learning, and deep learning models and the type of diverse data modalities used as behavioral questionnaires, neuroimaging data, eye-tracking signals, and even speech patterns. Furthermore, it highlights significant challenges about ML-based ASD systems, such as a limited data supply, imbalanced classes, and inability to generalize and constraints to the real-world clinical deployment.