Date of Award

Fall 12-2025

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

Program/Concentration

Computer Science

Committee Director

Sampath Jayarathna

Committee Member

Vikas G. Ashok

Committee Member

Faryaneh Poursardar

Abstract

Cognitive load provides insight into both how comfortably an individual is processing the information they are engaged with and the nature of their engagement, which is valuable insight to numerous fields. One method of estimating cognitive load is eye gaze measures which are acquirable in real-time through means adaptable to a variety of tasks and range of environments. Numerous advance gaze measures provide cognitive load insight but interpreting them requires expertise limiting the reach of eye gaze enabled cognitive load prediction. A potential solution to remove the expertise hurdle is through a simplified indicator of cognitive load that clearly conveys the level of cognitive load being experienced by an individual without the need to interpret a metric. This research investigates whether gaze measures are affected by cognitive load and whether those affects make it feasible to predict cognitive load using eye gaze measures. One analysis was training and testing machine learning models and performing feature ranking on gaze measures generated from a public dataset to predict unweighted NASA-TLX scores and NASA-TLX subscale scores in the dataset. The other analysis investigated whether the saccadic main sequence relationships are affected by cognitive load levels by calculating saccade gaze measures from three public datasets and testing if they satisfied the relationships or had any statistical differences between cognitive load levels. The machine learning models trained using support vector machine classifier and random forest performed with high F1 scores but struggled with low balanced accuracy, predicting unweighted NASA-TLX scores. However, the observed positive correlation between model performance and number of gaze measures used, high F1 scores, and lack of tuning and low data amount used led us to conclude gaze measure trained machine learning models are a feasible method for predicting cognitive load. The saccadic main sequence relationships were found to be statistically different between cognitive load levels and both relationships were violated in individuals experiencing cognitive load, making these violations feasible indications of cognitive load. This work provides a start for developing a gaze measure enabled simplified indicator of cognitive load by outlining two feasible methods for estimating cognitive load from gaze measures.

Rights

In Copyright. URI: http://rightsstatements.org/vocab/InC/1.0/ This Item is protected by copyright and/or related rights. You are free to use this Item in any way that is permitted by the copyright and related rights legislation that applies to your use. For other uses you need to obtain permission from the rights-holder(s).

DOI

10.25777/szed-r789

Additional Files
Owens supplementary files.zip (724 kB)

ORCID

0000-0002-0546-7350

Download

Available for download on Thursday, February 17, 2028

Share

COinS