AI-Powered Human-Computer Interaction Assisting Early Identification of Emotional and Facial Symptoms of Autism Spectrum Disorder in Children: "A Deep Learning-Based Enhanced Facial Feature Recognition System"

autism spectrum disorder (ASD) diagnosis often benefits from early identification. This paper proposes an enhanced deep learning model for analyzing facial features in children to aid in early ASD detection. The model integrates the YOLO architecture with convolutional neural networks (CNNs) employing transfer learning. Trained on a Kaggle dataset of 2,836 facial images (1,418 autistic and 1,418 non-autistic children), the model achieves high accuracy with MobileNet (95%), Xception (94%), and InceptionV3 (89%). This approach has the potential to improve existing detection methods for ASD and contribute to better human-computer interaction for individuals with special needs. Notably, the model demonstrates high specificity, effectively distinguishing autistic from non-autistic cases across various facial expressions in benchmark datasets.