Mental Workload Classification Using Short Duration EEG Data: an Ensemble Approach Based on Individual Channels

Abstract

Real-time mental workload assessment using EEG signals is an active research area that can potentially revolutionize the workplace for human operators. To make mental workload estimation practical in online every-day scenarios, it is vital to design classification models that can attain high accuracy when using short data fragments. In this study a subject-specific mental workload classifier has been designed that can classify two levels of mental workload with high accuracy when using data instances as short as 1.1s. The proposed model is an ensemble leaner that automatically extracts features of EEG channel data. Each base model of the ensemble classifier was specialized to learn spectral-temporal features of a single particular channel. Spatial information was added to the decision of the model by combining outputs of the channel-specific base classifiers through a majority voting method. The ensemble classifier was tested using different durations of data instances. Performance of the proposed ensemble classifier was better than a single classifier that used all channel data as input, for all instance durations tested. The proposed ensemble structure in this study can open new avenues to design ensemble learners based on different brain regions in EEG classification tasks.