Screening for dyslexia in children is a long procedure, which currently in Slovenia completely depends on the child's teachers and the pressure from parents. This is why the need for a computer system, which would be able to do a fast and easy screening, is increasing. In this thesis we analysed eye tracker data, which we acquired from an application for screening dyslexia. First, we identified fixations and saccades from raw data using the identification by velocity threshold (IVT) algorithm. The next step was to create multiple different groups of features, which describe the characteristics of eye movements. On these groups of features we used hierarchical clustering. Results showed that the best clustering was the one that used features, defined as averages over all tasks, but these clusters didn't differentiate well between dyslexics and non-dyslexics. By using clustering with features, which are defined as discriminatory for dyslexia in related works, we showed that in this clustering, we can find groups of children that are at higher risk of dyslexia.
|
