A Closer Look at Your Age: Determining Age Based on Eye Tracking Data

Abstract

Classifying the age of a person based on eye tracking data is something that has been successfully done for younger participants with data collected in a supervised setup in a laboratory. This thesis tries to find out if it is also possible to classify peoples age using gaze data acquired in a public environment without supervision. To do this, multiple supervised machine learning algorithms are trained and tested on this data. Their performances are evaluated using the accuracy, precision, recall & F1 score performance measures. Unfortunately, the performances of the classifiers are not great, mostly scoring an accuracy around the chance levels. However, the Support Vector Machine and the K-NN algorithm were able to achieve an accuracy of 20\% above chance on multiple classes, which shows that they are able to find patterns in the data. These are promising results and with extra features extracted from the data or with more advanced classifiers it might be possible to achieve high accuracy age classification based on unsupervised, publicly acquired data.