Gaze-Assisted Autism Spectrum Disorder Identification: A Fusion of Machine Learning and Deep Learning Approaches for Preemptive Identification
Judy Simon, Nellore Kapileswar, M Datchinamoorthi, S. Muthukumar, Keerthana Devi G
Abstract
This research proposes an all-encompassing method for diagnosing Autism Spectrum Disorder (ASD) by leveraging eye-tracking data and advanced machine learning techniques. In this study, gaze-tracking data was collected from individuals diagnosed with Autism Spectrum Disorder and those not exhibiting ASD characteristics as they participated in various tasks and stimuli. Principal Component Analysis was employed to streamline the eye-tracking data, retaining crucial information while enhancing computational efficiency. These refined features were then integrated into Convolutional Neural Networks as the primary modelling approach. CNNs excel in recognizing relevant spatial and temporal patterns within the visual data, leading to intricate feature extraction and classification. Additionally, we explored the efficacy of transfer learning by enhancing pre-trained CNN models with our eye-tracking dataset. This approach enables us to leverage knowledge from large-scale visual tasks, potentially enhancing the model's performance in ASD classification. The research findings highlight the success of this multifaceted approach in accurately categorizing individuals with ASD. The synergy of ML, CNN, and transfer learning models significantly enhances the accuracy of ASD diagnosis, offering a promising avenue for early identification and intervention. Consequently, this research has the potential to aid in the creation of accurate noninvasive instruments for diagnosing ASD, thereby accelerating timely assistance and treatment for individuals impacted by the condition.