Autism Detection via Eye Tracking

abstract

A non-invasive machine-learning approach to autism spectrum screening that classifies subjects from eye-tracking data, using gaze patterns and fixation duration as behavioural biomarkers rather than clinician-administered instruments.

my contribution

Built the two-stage pipeline — ResNet18 for deep feature extraction from eye-tracking frames, followed by Random Forest classification — and ran the evaluation across precision, recall, and F1 to establish clinical reliability rather than reporting accuracy alone.

keywords

• Eye tracking
• Autism screening
• ResNet18
• Random Forest
• Behavioural biomarkers