Utilizing Artificial Intelligence to Diagnose Autism Spectrum Disorder Based on Eye Tracking Saccades

Autism Spectrum Disorder (ASD) refers to a group of neurodevelopmental disorders, usually characterized by impairments in social interaction skills, struggle with verbal and nonverbal communication, and repetitive behaviors. ASD can not be diagnosed with any one medical test and presents a wide range of symptoms that make it hard to identify in patients of all ages and severities. This paper proposes a novel method of diagnosis to shorten time, resources, and uncertainty during this process. This predictive machine learning model uses eye tracking (ET) and follows saccadic eye movements in patients to identify patterns, irregularities, and fixation time as symptoms of ASD. The results of experimental testing show that the Two Class Boosted Decision Tree is the most accurate model to use, with an estimated accuracy score of 88–91%. As presented by the model, the tracking of sensorimotor activity such as saccades and application into machine learning has the capability to improve the diagnosis of patients across the spectrum.