Classification of Children With Autism and Typical Development Using Eye-Tracking Data From Face-to-Face...

Area of Science:

• Neuroscience
• Psychology
• Computer Science

Background:

• Previous studies utilized machine learning (ML) with eye-tracking data from image viewing for autism spectrum disorder (ASD) identification.
• Gaze behavior differs between real-world interactions and image-viewing tasks.
• No prior research explored eye-tracking data from face-to-face conversations for ASD detection.

Purpose of the Study:

• To determine if eye-tracking data from face-to-face conversations can classify children with ASD and typical development (TD).
• To investigate if combining visual fixation and conversation length features enhances classification accuracy.

Main Methods:

• Eye tracking was conducted on children with ASD and TD during face-to-face conversations.
• Four ML classifiers (SVM, linear discriminant analysis, decision tree, random forest) were employed.
• Forward feature selection identified optimal features for classification.

Main Results:

• A maximum classification accuracy of 92.31% was achieved using a support vector machine (SVM) classifier.
• Combining visual fixation and session length features yielded higher accuracy than individual features alone.
• Visual fixation and session length features individually achieved a maximum accuracy of 84.62%.

Conclusions:

• Eye-tracking data from face-to-face conversations can accurately classify children with ASD and TD.
• This suggests potential for objective ASD screening in everyday social interactions.
• Future research requires larger, diverse samples and multimodal data for validation.