Enhancing mental workload recognition: a comparison of complexity-based eye movement metrics and conventional...

Area of Science:

• Human-Computer Interaction
• Cognitive Science
• Biomedical Engineering

Background:

• Conventional eye-movement metrics show limitations in recognizing mental workload (MWL) due to inconsistent dynamic time-series pattern capture.
• Complexity-based features are emerging as potentially more robust indicators of MWL.

Purpose of the Study:

• To evaluate the effectiveness of complexity-based eye-movement features, incorporating intrinsic mode functions (IMFs), as indicators of MWL.
• To compare the performance of complexity-based features against conventional metrics for MWL recognition.

Main Methods:

• Participants performed mathematical tasks of varying MWL, with eye movements recorded and NASA-RTLX assessments conducted.
• Eye-movement data were decomposed using empirical mode decomposition, and multiscale entropy was computed.
• Machine learning models were trained using both conventional and complexity-based feature sets.

Main Results:

• Complexity-based features demonstrated more consistent capture of MWL effects compared to conventional metrics.
• Classification accuracy for subjective MWL recognition was higher with complexity-based features (68% vs. 53%).
• Classification accuracy for task-based MWL recognition was also higher with complexity-based features (73% vs. 57%), showing a 15-16% improvement.

Conclusions:

• Complexity-based eye-movement metrics, particularly those derived from IMFs, offer superior performance for MWL recognition.
• These findings support the integration of complexity-based metrics to enhance the accuracy and reliability of human-computer interaction systems.