Classifying Sex Based On Eye Tracking Data: A Machine Learning Study

Abstract

This paper describes a study into the classification of gender based on viewing behavior. This was done with the data of 1242 visitors of the NEMO museum, to which we had to pick a classification algorithm and decide on what features to use with this algorithm to train and test our given data with. We evaluated the algorithm based on multiple machine learning measures, such as Precision, Recall and F1-score, but the most important measure, which was also the measure we were basing our evaluation on, was the Accuracy measure. Our criteria for a good algorithm was set to 70%, which was based on related work. Our algorithm with the implemented feature set got exactly that as Accuracy, to which we can conclude that it is indeed possible to program an algorithm that can correctly classify sex based on eye tracking data. This has a few implications: by further analysing eye tracking data and successfully furthering algorithms to also correctly classify variables such as age and mood of a person, we can predict the way people are going to behave and make things such as advertisements more effective.