Application of machine learning on eye-tracking data for autism detection: The case of high-functioning adults

Autism Spectrum Disorder (ASD) is a heterogeneous neurodevelopmental disorder which, when diagnosed early, facilitates early intervention. However, the lack of objective assessment tools as well as the difficulty to administer them to children under five retards autism detection and subsequent intervention. High-functioning autism individuals often exhibit functional deficits, not readily apparent, which deteriorate their quality of life and that of their families. The employment of technologies such as Machine Learning (ML) and Robotics holds promises regarding objective autism assessment and high-functioning assessment in particular. In this study, a dataset gathered in a previous study was utilised aiming to classify high-functioning autistic adults engaged in web-browsing tasks: a Browse and a Search one. Various machine learning algorithms were implemented resulting in increased high-functioning autism classification accuracy, i.e., 71.9% and 71.4% for the Browse and Search tasks respectively. This methodology could enhance autism research regarding increased autism classification accuracy and objective and less expensive autism assessment. The limitations encountered and future scope for research are presented, as well.