EEG-based Cross-Subject Workload Classification

EEG (Electroencephalography) data has been utilized to discriminative levels of mental workload. Generally speaking, there are three kinds of workload classifiers:

1. Subject-specific Classifier: trained and tested on the same subject
2. Cross-subject Classifier: trained on group of subjects and tested on the same group of sbjects
3. Subject-independet Classifier: trianed on a group of subjects while tested on a novel subject

Although most of the existing works have achieved great success in classifying workload for individual subjects, it remains a significant challenging problem for us to construct a classifier for a group of subjects or even a totally novel subject due to the tremendous amount of variations across subjects.

We are working to build a cross-subjet classifier and a subject-independent classifier. Currently, Neural Network and SVM are the most popular classifiers used for workload classification. However we propose to utilize graphical models, probabilistic theories and the Bayesian ideas for learning and inference.

So far, we have successfully established a cross-subject classifier that can achieve even better performance than the subject-specific classifier by introducing a hidden node to a Naive Bayes classifier. It further evolved to a hierarchical structure that can impose higher domain knowledge on the parameters.

Our future work focuses on building a totally subject-independent classifier.