Facial-Based Autism Classification Using Support Vector Machine Method

Abstract

Autism Spectrum Disorder (ASD) is a complex neural developmental condition characterized by difficulties in communication, social interaction, and delayed brain development. Despite previous studies, there is a need to explore and enhance autism classification techniques using facial data. This research aims to classify individuals with autism based on facial images using the Support Vector Machine (SVM) method. It also evaluates the performance of SVM-based classification with HOG and SURF feature extraction, contributing to the identification of autism through facial features. A dataset of 200 facial images of students, including individuals with and without autism, was analyzed. The data was divided into 80:20 and 70:30 splits for training and testing purposes. SVM models with HOG and SURF feature extractions were evaluated using accuracy, precision, recall, and F1-Score metrics. The HOG-SVM and SURF-SVM models showed consistent performance in both data splitting scenarios. Accuracy values exceeded 0.88, and precision, recall, and F1-Score values were above 0.9. The 80:20 data split demonstrated improved performance, especially for the HOG-SVM model. Both HOG and SURF feature extraction methods showed good performance in classifying autism data. The SVM model with HOG achieved an accuracy of 0.95 in the 80:20 data split, while the SURF model achieved 0.9. Early autism detection based on facial data holds potential for use in student selection in elementary schools. However, the study has limitations due to limited data and the focus on accuracy alone. Future research can expand the data size, explore other feature extraction methods, and implement advanced deep learning techniques to improve classification performance and contribute further to autism detection based on facial data .