Facial Image-Based Autism Detection: A Comparative Study Among Machine Learning, Deep Learning and Transfer Learning

For prompt intervention and better results, early autism screening is essential. This work investigates the use of face image databases for autism classification in a non-invasive, cost-effective manner. Support Vector Machine (SVM), Convolutional Neural Network (CNN), and EfficientNet were the three classification models used. With the greatest accuracy of 85.67% and superior performance across important performance parameters including precision, recall, and F1-score, EfficientNet surpassed the competition. The work demonstrates how EfficientNet may be used to detect minute face traits associated with autism, providing a good substitute for more traditional techniques including brain signal, imaging, and video analysis. This study highlights the potential of cutting-edge machine learning models for autism identification and promotes more investigation into bigger datasets to improve diagnostic accessibility and accuracy.