Characterising Eye Movement Events With an Unsupervised Hidden Markov Model

Abstract

Eye-tracking allows researchers to infer cognitive processes from eye movements that areclassified into distinct events. Parsing the events is typically done by algorithms. Previousalgorithms have successfully used hidden Markov models (HMMs) for classification but canstill be improved in several aspects. To address these aspects, we developedgazeHMM, analgorithm that uses an HMM as a generative model, has no critical parameters to be set byusers, and does not require human coded data as input. The algorithm classifies gaze datainto fixations, saccades, and optionally postsaccadic oscillations and smooth pursuits. Weevaluated gazeHMM’s performance in a simulation study, showing that it successfullyrecovered HMM parameters and hidden states. Parameters were less well recovered whenwe included a smooth pursuit state and/or added even small noise to simulated data. Weapplied generative models with different numbers of events to benchmark data. Comparingthem indicated that HMMs with more events than expected had most likely generated thedata. We also applied the full algorithm to benchmark data and assessed its similarity tohuman coding. For static stimuli, gazeHMM showed high similarity and outperformedother algorithms in this regard. For dynamic stimuli, gazeHMM tended to rapidly switchbetween fixations and smooth pursuits but still displayed higher similarity than otheralgorithms. Concluding that gazeHMM can be used in practice, we recommend parsingsmooth pursuits only for exploratory purposes. Future HMM algorithms could usecovariates to better capture eye movement processes and explicitly model event durationsto classify smooth pursuits more accurately.

Keywords

• eye tracking
• eye movements
• event detection
• dependent mixture models
• saccades
• smooth pursuit