An Intelligent Deep Learning Approach to Recognize Autism Spectrum Disorder using Hybrid Optimization

Abstract

Introduction: This study analyzes the "Visualization of Eye-movement Scanpaths in ASD" dataset. There are clear eye-tracking traces images of ASD in this dataset. ASD is a disorder of development. It creates issues with the manner in which people behave, communicate, and interact [5]. Eye-tracking informs us about ASD. This is because abnormal eye movement patterns are a major sign of ASD. The dataset informs us about the way eyes move. It informs us how people with and without ASD view things. We enhanced our sorting with the use of advanced image analysis techniques. We applied deep learning and machine learning models in our study. We employed data augmentation techniques to expand and enhance our dataset. This boosted our models. We tried various Machine Learning (ML) and Deep Learning (DL) techniques. They were Decision Tree, Logistic Regression and Random Forests. We also implemented deep learning models such as MobileNetV3Large. We tested and trained them. Finally, MobileNetV3Large was the winner. It's tiny but effective deep learning model. It accurately labeled ASD and non-ASD groups 90.48% of the time. The study proves the capability to use eye-movement scanpaths and ML algorithms in the detection and investigation of Autism in the early stage more effectively. The technique provides a great and painless way of screening ASD from visual patterns in eye-tracking. This is most likely crucial in early diagnosis and action towards persons suffering from the disorder.