SCREENING FOR DYSLEXIA USING EYE TRACKING

Dyslexia is a neuro developmental reading disorder that degrades the speed and accuracy of word recogni- tion, and as a consequence, impedes reading fluency and text comprehension. Between 5 and 10 percent of the population are normally affected by it. It has long been known that the eye movements of dyslexic readers differ from those of typical readers. The dataset for this study has been taken from the dataset used by a similar study (Benfatto et al., 2016). The experiments reported by the authors are based on eye tracking data from 185 subjects participating in the Kronoberg reading development project, a longitudinal research project on reading development and reading disability in Swedish school children running between 1989 and 2010. For our present study, we use eye movement recordings made while the subjects were reading a short natural passage of text adapted to their age. Recordings were available for 185 subjects, 97 High Risk (HR) subjects (76 males and 21 females) and 88 Low Risk(LR) subjects (69 males and 19 females Machine learning based predictive model developed in this study employ feature set based on eye fixations and saccades parameters and can be used to give individual level diagnosis with high sensitivity and specificity. Using statistical cross-validation techniques on a sample of 97 dyslexic and 88 control subjects, we achieve a classification accuracy of over 96% with balanced levels of sensitivity and specificity. Diagnostic follow-up of a screening result is always necessary so that intervention strategies can be personalized. Nevertheless, early identification of individuals in need of support is the first important step in this process and using eye tracking along with this system during reading may prove very useful. The system's accuracy can be further enhanced by collecting a larger sample and then training these and other classification models.