EEG workload prediction in a closed-loop learning environment

Abstract

The issues of developing an online EEG-based adaptive learning environment are examined in this thesis. The aim is to adapt instructional learning material in real-time, to support learners in their individual learning process and keep them in their optimal workload capacity range during learning. First, suitable learning material is designed, which does not cause artifacts and induces confounds in the EEG data. Second, the most suitable features for an online workload detection in EEG data are determined, by using a variety of pre-processing and feature selection methods, as connectivity and independent component analysis. Third, generalizable classification methods like cross-task classification and cross-subject regression are developed, to enable a workload prediction across a variety of tasks, independently from subjects. In an offline analysis, the cross-subject regression leads to a higher workload prediction accuracy as the cross-task classification. Since the workload prediction across subjects is more precise, this method is used for the subsequent online study. Therefore, the achieved findings and developed classification methods will finally be applied in an online study. The difficulty level of the presented learning material is adapted in real-time, dependent on the predicted workload of each subject. Furthermore, the applicability and efficiency of an online EEG-based adaptive learning environment is investigated and assessed. Comparing the EEG-based learning environment with an error-adaptive learning system, which is state of the art, the induced learning effects are similar. Thus, the learners can successfully be supported in their individual learning process using an EEG-based adaptation of the learning material, by keeping them in their optimal workload range for learning.