Advanced Hybrid Transfer Learning Approaches for Autism Spectrum Disorder Detection Using Facial Features

Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition characterized by deficits in social interactions, communication, and emotional regulation. Early identification is crucial for improving outcomes. This work presents a DL approach for ASD detection using facial features extracted from the Autistic Children Facial Dataset. Initially, transfer learning models including VGG16, EfficientNet, DenseNet, and MobileNet were applied individually to extract and analyze critical facial characteristics. VGG16 and DenseNet provided detailed high-level features, EfficientNet offered efficient feature extraction, and MobileNetV3 enabled rapid processing. Later, a hybrid model was developed by merging the features extracted from the above applied models for emotion-based ASD identification. The hybrid approach was trained and evaluated using SVM and random forest ML models. Experimental results demonstrated that the hybrid approach enhanced detection accuracy for early ASD detection.