Machine Learning Approaches for Autism Spectrum Disorder Detection: A Systematic Review of Age-Specific Applications and Performance Metrics

Autism Spectrum Disorder is one of the biggest concerns in the healthcare sector, and it’s crucial to diagnose it at an early stage for patients with Autism Spectrum Disorder. This review focuses on the use of machine learning in diagnosing Autism Spectrum Disorder, drawing data from 100 papers between 2015 and 2024. We touched every possible method starting from the classic ones like Support Vector Machines (SVMs) to the new ones like federated learning. Proving the federated learning is actually great since it is very precise (up to 98%) while keeping people’s information personal, which is a crucial matter in the healthcare industry. But one cannot write-off the basic framework where people use standard machine learning models such as SVMs, which at this point achieve around 92% accuracy. Also, they are more convenient to be implemented in small clinics that do not possess many great computers, and etcetera. This review suggests that the most suitable ML approaches for Autism Spectrum Disorder detection need to consider accuracy, privacy and availability of resources. Lately, more developed technologies provide even better outcomes; nevertheless, conventional techniques provide terrific options for clinics without much complicated systems available. Thus, the study offers meaningful suggestions to facilitate the choice of the most suitable methods based on the comparison between these approaches. In sum, this review spans the existing gap between research advancements in state-of-art machine learning techniques and practical healthcare settings and provides important recommendations for enhancing Autism Spectrum Disorder screening across various contexts.