EEG-based workload estimation across affective contexts.

Area of Science:

• Neuroscience
• Human-Computer Interaction
• Signal Processing

Background:

• Workload estimation using electroencephalographic signals (EEG) is crucial for adaptive human-computer interaction.
• Real-world systems require robustness against contextual changes, such as user mood or stress.
• Existing EEG-based workload classifiers need evaluation for their resilience to affective context shifts.

Purpose of the Study:

• To investigate the resilience of state-of-the-art EEG workload classification against stress.
• To develop and validate a novel experimental protocol for manipulating affective context.
• To assess the generalization capability of workload classifiers across stressful and non-stressful contexts.

Main Methods:

• A novel protocol manipulated affective context (stressful/non-stressful) during a two-level workload task.
• Recorded self-ratings, behavior, and physiological data from 24 participants.
• Tested subject-specific workload classifiers using frequency-domain, time-domain, or combined features for cross-context generalization.

Main Results:

• EEG workload classifiers demonstrated transferability between affective contexts, albeit with reduced performance.
• Performance degradation was observed regardless of the feature domain (time or frequency).
• Cross-context training emerged as an effective method to enhance feature resilience against task-unrelated variations.

Conclusions:

• Cross-context training significantly improves the robustness of EEG workload classifiers in varying affective states.
• Frequency-domain features, when trained across contexts, achieved performance comparable to within-context training.
• Findings are vital for developing reliable passive brain-computer interfaces and neurophysiology-based workload detection systems.