Authors - Alyssa C. Vicente, Cedirick Santiago, Elmer M. Alino, Ma. Yvonne Czarina C.Angcaya, Benedict G. Bautista, St. Joseph M. Lumbog Abstract - This systematic review investigates the application of machine learning (ML) and deep learning (DL) in the early detection of Autism Spectrum Disorder (ASD), a neurodevelopmental condition characterized by social and communication deficits. Adhering to the 24-step framework by Muka et al. and PRISMA 2020 guidelines, the methodology involved a rigorous search of four academic databases—IEEE Xplore, Scopus, PubMed, and ACM Digital Library— identifying 67 records. Ultimately, 10 peer-reviewed studies published between 2020 and 2024 were analyzed based on their use of real-world datasets and quantitative metrics. Results indicate that ML models, particularly Convolutional Neural Networks (CNNs) and ensemble classifiers, achieve high predictive performance with accuracies between 80% and 94%. The findings highlight that behavioral data from home videos and eye-tracking scan paths serve as effective indicators for remote, scalable screening. However, the review identifies significant gaps, including small, homogeneous datasets and a lack of model interpretability. To advance the field, future research must focus on Explainable AI (XAI), multimodal fusion, and the development of large-scale, multicultural, open-access datasets to ensure clinical trust and global generalizability.
Open Zoom