Machine learning for early detection of autism spectrum disorder: a data-driven approach to public health support

Aim This systematic review aimed to evaluate how machine learning (ML) models have been utilized in the early detection of autism spectrum disorder (ASD) among children aged 0–12, focusing on model types, data modalities, diagnostic performance, implementation feasibility, and ethical considerations.Methods Following PRISMA guidelines, a comprehensive search was conducted across six databases (2010–2025). Studies were included if they employed ML for ASD detection in young children using clinical, behavioral, imaging, or sensor-based data. Data extraction and quality assessment (using PROBAST) were conducted independently by two reviewers. Narrative synthesis was applied due to methodological heterogeneity.Results Thirty-six studies met the inclusion criteria. Supervised ML algorithms such as support vector machines (SVMs), random forests, and deep neural networks demonstrated high diagnostic accuracy (AUCs up to 0.98), particularly when using multimodal input (e.g. EEG + behavioral checklists). However, only a minority of studies used external validation. Real-world implementation was limited, and ethical issues (e.g. data privacy, algorithmic bias) were inconsistently addressed. Reproducibility was also hindered by incomplete reporting.Conclusions ML shows significant promise for early ASD detection, especially through multimodal models integrated into clinical workflows. Nonetheless, widespread adoption requires addressing validation gaps, transparency, ethical safeguards, and clinician training. Future research should emphasize real-world evaluation and equity-driven AI deployment.