An Intelligent Assistive System for Autistic Learners

Abstract

This study presents the design and implementation of an educational assistive system tailored for individuals with Autism Spectrum Disorder (ASD), leveraging eye-tracking technology and machine learning algorithms. The objective is to support communication and learning experiences through ocular gaze and fixation-based keyboard selection. An existing collection of scanpath images obtained from 15 children with ASD was utilized, and a customized heterogeneous dataset was developed by extracting fixation and gaze-related features using machine learning-based processing techniques. In total, 320 scanpath images were analyzed to capture fixation points and gaze directions during left–right and up–down navigation tasks. Based on fixation metrics and Area of Interest (AOI) analysis, a novel keyboard selection method was proposed, distinct from traditional digital keyboards. The system was evaluated using an optimized ensemble algorithm, achieving an accuracy of 83.44%, with macro-averaged F1, precision, and recall of 80.43%, 81.43%, and 80.57%, respectively, outperforming the baseline models. The results indicate the system's potential to improve communication efficiency and usability, suggesting its suitability for facilitating hands-free keyboard control. This assistive system is intended to support interaction with parents, teachers, and caregivers, which may help promote social engagement for individuals with ASD. Overall, this research demonstrates the promise of integrating eye-tracking and machine learning in educational tools, while contributing a customized dataset and a novel selection method to support neurodiverse learners. The associated code, AOI layout files, and a sample of the derived dataset regarding this study are publicly available to enable independent replication of the analysis. These materials can be accessed at https://github.com/AfrinSadi/AssistiveAI-Autism.git.