Predictions of task using neural modeling.

Area of Science:

• Neuroscience
• Cognitive Science
• Human-Computer Interaction

Background:

• Brain-computer interfaces (BCIs) passively assess brain activity to predict user states and optimize human-machine team (HMT) performance.
• Neural features generalize differently across contexts, experimental manipulations, and sessions, posing a challenge for neuroergonomics.
• Predicting cognitive workload, a key psychological construct, is crucial for adaptive HMT systems.

Purpose of the Study:

• To quantify the generalization of neural features and modeling approaches for predicting cognitive workload under various task manipulations.
• To evaluate the performance of different machine learning models in predicting single- and multi-task cognitive workload.
• To assess the impact of within-session versus between-session predictions on model accuracy.

Main Methods:

• Trained and tested 20 support vector machine (SVM) models using distinct subsets of spectro-temporal neural features.
• Evaluated model accuracy in predicting task type (monitoring, communications, tracking) and quantity (one, two, or three) simultaneously.
• Investigated within-session and between-session prediction accuracy at the individual level.

Main Results:

• Gamma activity across all recording locations consistently outperformed other neural feature subsets.
• Modelers must consider diverse manipulations that influence the same underlying psychological construct (cognitive workload).
• Model accuracy decreased when using mobile EEG systems with fewer electrodes.

Conclusions:

• Gamma activity is a robust neural feature for predicting cognitive workload in BCI applications.
• The study provides a practical modeling solution for predicting task states via brain activity.
• Findings offer insights into the trade-offs between model accuracy and the practicality of mobile EEG deployment.