ATTENTIONAL MODEL FOR DETECTING ATTENTION IN CHILDREN WITH AUTISM SPECTRUM DISORDER

Abstract

Background: Children with Autism Spectrum Disorder (ASD) exhibit an attentional deficit during learning tasks. Thus, teachers and caregivers need to identify when to give attentional support. Still, it is challenging to tell when they are inattentive. Aim: This research investigates the attentional behaviors peculiar to children with ASD in order to develop an attentional model. Method: The three main methods used to achieve the research aims include exploratory, experimental, and machine learning model development. The exploratory study consists of a systematic literature review on attention assessment and semi-structured interviews with 17 experts in the ASD field. In the experimental study, 46 children (ASD n=20) and typically developing (TD n=26) took four different attention tasks. At the same time, we track their eye and face in real-time. The generated data were pre-processed and fed into Support Vector Machine to classify attention and inattention during learning. We compared the performance of three model types: face-based, gaze-based, and hybrid-based. Results: The analysis of exploratory data revealed that face and eye-tracking are crucial for measuring attention. The experimental data demonstrated the potential of face and eye-tracking for detecting attention and evaluating ASD. The attentional model also revealed that the participant-dependent model has higher predictive power than the participant-independent model for the ASD group. The three attentional models, face-based (AUC= 0.957), gaze-based (AUC= 0.998) and hybrid-based (AUC= 0.996) show the potential of detecting attention of children with ASD. Conclusion: This study showed that children with ASD exhibit attentional behavior differently with a unique face and eye-tracking features. These findings stress the direction of model design from traditional one-size-fits-all models to personalized models, especially in children with ASD.

Date of Award	2021
Original language	American English
Awarding Institution	HBKU College of Science and Engineering