Task-generic mental fatigue recognition based on neurophysiological signals and dynamical deep extreme learning machine

The electroencephalography (EEG) based machine-learning model for mental fatigue recognition can evaluate the reliability of the human operator performance. The task-generic model is particularly important since the time cost for preparing the task-specific training EEG dataset is avoid. This study develops a novel mental fatigue classifier, dynamical deep extreme learning machine (DD-ELM), to adapt the variation of the EEG feature distributions across two mental tasks. Different from the static deep learning approaches, DD-ELM iteratively updates the shallow weights at multiple time steps during the testing stage. The proposed method incorporates the both of the merits from the deep network for EEG feature abstraction and the ELM autoencoder for fast weight recompuation. The feasibility of the DD-ELM is validated by investigating EEG datasets recorded under two paradigms of AutoCAMS human-machine tasks. The accuracy comparison indicates the new classifier significantly outperforms several state-of-the-art mental fatigue estimators. By examining the CPU time, the computational burden of the DD-ELM is also acceptable for high-dimensional EEG features.