Eye Tracking Data Clustering

TONY LIN

Exchange Summary

A large, potential feature of psychological data is to know where humans tend to look in an experiment and how often, namely their areas of interests or AOIs. Unfortunately, currently many AOIs are hardcoded. When setting up an eye tracking experiemnt, one usually measures and dictates what is an AOI. This could lead to many errors since AOIs have a hard separation; datapoints that convey an AOI end up falling outside of this classification. Thus, two approaches for semi-automated AOI detections are presented, namely a direct GMM method and iterative KMeans with GMM classification. Both are unsupervised learning and provide reasonable results with noticeable differences between the two depending on the data. The library also provides a number of filtering and labeling methods on eye tracking data. By utilizing databased approaches to AOIs, one can gain a far more accurate model and thus, learn a lot more about human behavior in a variety of situations.