Comprehensive Analysis of Machine Learning Techniques in Autism Spectrum Disorder Detection

This review investigates the application of machine learning for (ASD) Autism Spectrum Disorder detection by reviewing existing models, discussing challenges, identifying future pathways for increasing detection and intervention services. We conducted a systematic review of studies from 2015 to 2024 that use ML techniques, including analysis of deep learning, traditional algorithms, and hybrid models, for the ASD detection, involving analyses of neuroimaging, behavioral, genetic, and physiological data. CNNs excel in neuroimaging, while RNNs aid time series behavioural data. Overall, the mutlimodal approach was more accurate that models reliant on single sources, ranging from 85% to 98% accurate. The two main issues were heterogeneity of data and limited feature selection. There is evidence in the literature, namely, that ML techniques may enhance ASD monitoring and identification. The future work should focus on creating datasets that consistently contain people from various backgrounds, creating interpretable machine learning models, and conducting clinical on these techniques.