Can artificial intelligence and face recognition using deep learning detect emotions in children with autism?

Background/Objectives This study aimed to evaluate the performance of deep learning models for recognizing facial expressions of children with autism through face recognition technologies. Methods Conducted between November 2024 and February 2025, this research introduced a novel deep learning architecture, AutismEfficientNet, by integrating EfficientNetV2 and MobileNetV3. Two previously published datasets—Dr. Fatma M. Talaat’s Emotions of Autistic Children and the FERAC dataset—were used, both containing labeled facial images of children with autism displaying various emotions. Images were preprocessed and divided into training, validation, and test sets. Emotion classification was performed using convolutional neural networks. Models were evaluated using accuracy, sensitivity, specificity, and F1 scores. Results Deep learning models were able to accurately recognize emotional expressions in children with autism. The proposed AutismEfficientNet achieved the highest accuracy (75.8%) among the tested models, outperforming EfficientNetV2-L (72.4%) and MobileNetV3-L (70.2%). Conclusions The findings indicate that AutismEfficientNet provides improved classification accuracy compared to standard architectures when applied to datasets of children with autism. Further validation is required in clinical settings and diverse populations before integration into pediatric care.