Interpretable Machine Learning Models for Three-Way Classification of Cognitive Workload Levels for Eye-Tracking...

Area of Science:

• Neuroscience
• Cognitive Psychology
• Computer Science

Background:

• Cognitive workload assessment is crucial for understanding mental fatigue and cognitive processes.
• Eye-tracking data offers a non-invasive method for monitoring cognitive states.
• The Digit Symbol Substitution Test (DSST) is a common tool for evaluating cognitive function.

Purpose of the Study:

• To assess cognitive workload levels using machine learning models.
• To investigate the effectiveness of interpretable machine learning in cognitive workload analysis.
• To identify key features contributing to cognitive workload classification.

Main Methods:

• Collected eye-tracking data from 29 healthy volunteers performing three versions of the DSST.
• Developed and analyzed eight three-class classification machine learning models.
• Applied interpretable machine learning techniques to determine feature importance.

Main Results:

• Machine learning models achieved high classification performance, with the best F1 score reaching 0.95 on the complete feature set.
• Interpretable machine learning enhanced classification accuracy to 0.97 using only seven of 20 features.
• Feature importance analysis provided insights into brain cognitive functions related to workload.

Conclusions:

• Machine learning, particularly interpretable models, is effective for assessing cognitive workload using eye-tracking data.
• Reducing feature sets based on importance analysis can improve classification performance and efficiency.
• This approach offers valuable insights into cognitive processes and mental fatigue.