Eye Tracking-Based Diagnosis and Early Detection of Autism Spectrum Disorder Using Hybrid Algorithm

The aim of the research is to evaluate how eye monitoring technology works with machine learning and deep learning methods to detect autistic spectrum disorder (ASD) at an early stage. Research uses various ML and DL approaches that include additional neural networks (FFNN) and artificial neuron networks (Anns) and also include a hybrid algorithm for benchmark classification. The levels of high accuracy of machine learning methods arise from the use of the algorithms of the local binary pattern (LBP) and the matrix of the gray stages (GLCM) in the categorization of elements for the neuron network raw materials (FFNN) and artificial neural networks (ANNS). The application of pre-trained convolutional neural networks (CNN) models Googlenenet and Resnet-18 by means of deep learning methods achieved accuracy of functional maps based on 0.94% and 0.99% extraction. The hybrid technique used deep learning models along with machine learning components through a combination of GoogleNet and ResNet-18 and Support Vector Machines achieved high diagnostic accuracy reaching rates of 0.98% for GoogleNet + SVM and 0.96% for ResNet-18 + SVM. The prediction accuracy of ASD remains high based on the hybrid Approach which demonstrates comparable results to various other prediction models. AI analysis of eye-tracking data has shown promising features for exact and early ASD identification in this research because it offers vital opportunities to boost clinical evaluation and therapy approaches.