Feasibility of Machine Learning in Predicting Features Related to Congenital Nystagmus

Congenital nystagmus is an ocular-motor disease affecting people’s visual acuity since their first years of life. Electrooculography is used to perform eye tracking in these patients, giving the possibility to extract a wide variety of parameters. The relationships among all these variables were analysed in the past and the aim of this paper is to perform a new analysis employing more recent techniques, those of machine learning. The electrooculography of 20 patients was recorded, signals were pre-processed, and some parameters were extracted through a custom-made software. Knime analytics platform was chosen in order to build predictive models using Random Forests and Logistic Regression Tree algorithms and some evaluation metrics were computed. The visual acuity and the variability of eye positioning were predicted employing five and six variables, respectively. In terms of coefficient of determination, visual acuity had values over 0.72 and variability of eye positioning over 0.70. Compared to the results obtained without machine learning algorithms during the past years, these values become more valuable. In conclusion, this approach showed its feasibility in detecting relationships among variables related to congenital nystagmus; it could be tested in order to find new and stronger relationships among these variables and be of support for clinicians.