Early Stage Detection of Autism Spectrum Disorder

Abstract

Autism Spectrum Disorder (ASD) is a developmental condition impacting communication, behavior, and social interaction, and it affects approximately 2% of children in the U.S. Early identification and intervention are critical to improving long-term outcomes for children with ASD, yet delays in diagnosis remain a significant barrier. Current screening tools, such as the Modified Checklist for Autism in Toddlers, revised with Follow-Up (M-CHAT-R/F), are widely used and effective in identifying potential ASD cases. However, these tools often exhibit limitations in specificity, leading to a high rate of false positives and a lower positive predictive value. This poses challenges for healthcare systems and families, as unnecessary follow-ups and evaluations can strain resources and cause undue stress.To address these limitations, a machine learning-based framework has been proposed to enhance the accuracy of early ASD detection. By leveraging advanced computational techniques, the framework evaluates the performance of eight different algorithms, including AdaBoost, Random Forest, and Support Vector Machines (SVM). A voting classifier, which combines predictions from multiple models, is also employed to improve the robustness and reliability of probability assessments. In addition to traditional machine learning methods, the study explores the potential of image-based approaches, such as Convolutional Neural Networks (CNNs) and Deep Neural Networks (DNNs). These techniques could provide valuable insights when integrated with behavioral data, offering a more comprehensive diagnostic tool. Furthermore, advancements in motion analysis and the use of biomarkers are discussed as complementary strategies to refine the screening process. However, while these technologies show promise, further research and validation are needed to ensure their effectiveness and applicability across diverse populations