Deep Learning Based Autism Detection in Children from Facial Features

Abstract

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that affects a child’s ability to  communicate, interact socially, and display normal behavioral patterns. Early detection of ASD plays a crucial role in providing timely therapy and improving the developmental outcomes of affected children. However, traditional diagnostic prsocedures depend heavily on behavioral observation and clinical expertise, which are time-consuming, expensive, and often unavailable in many regions. This research proposes a deep learning-based automated system for identifying Autism Spectrum Disorder(ASD) in children using facial images. The proposed system utilizes Convolutional Neural Networks (CNN) and transfer learning models such as VGG16, ResNet, and MobileNet for automatic extraction of facial features and classification. Image preprocessing techniques such as face detection, cropping, resizing, and normalization are applied to improve image quality and enhance model performance. Data augmentation techniques are also employed to increase dataset diversity and reduce overfitting. The system classifies input images into ASD or Non-ASD categories and provides a prediction confidence score. Experimental results demonstrate that deep learning-based facial feature analysis can support early ASD screening with improved accuracy and efficiency. The proposed system aims to provide a cost-effective, non-invasive, and user-friendly screening tool that can assist parents, healthcare professionals, and early intervention programs.