Classifying Mental States From Eye Movements During Scene Viewing

Keywords

Classification; Eye movements; Linear discriminant; Multivariate analysis; Scene viewing

Abstract

How eye movements reflect underlying cognitive processes during scene viewing has been a topic of considerable theoretical interest. In this study, we used eye-movement features and their distributions over time to successfully classify mental states as indexed by the behavioral task performed by participants. We recorded eye movements from 72 participants performing 3 scene-viewing tasks: visual search, scene memorization, and aesthetic preference. To classify these tasks, we used statistical features (mean, standard deviation, and skewness) of fixation durations and saccade amplitudes, as well as the total number of fixations. The same set of visual stimuli was used in all tasks to exclude the possibility that different salient scene features influenced eye movements across tasks. All of the tested classification algorithms were successful in predicting the task within a single participant. The linear discriminant algorithm was also successful in predicting the task for each participant when the training data came from other participants, suggesting some generalizability across participants. The number of fixations contributed most to task classification; however, the remaining features and, in particular, their covariance provided important task-specific information. These results provide evidence on how participants perform different visual tasks. In the visual search task, for example, participants exhibited more variance and skewness in fixation durations and saccade amplitudes, but also showed heightened correlation between fixation durations and the variance in fixation durations. In summary, these results point to the possibility that eye-movement features and their distributional properties can be used to classify mental states both within and across individuals.

Publication Date

12-1-2015

Publication Title

Journal of Experimental Psychology: Human Perception and Performance

Volume

41

Issue

6

Number of Pages

1502-1514

Document Type

Article

Personal Identifier

scopus

DOI Link

https://doi.org/10.1037/a0039673

Socpus ID

85027938658 (Scopus)

Source API URL

https://api.elsevier.com/content/abstract/scopus_id/85027938658

This document is currently not available here.

Share

COinS