Mental workload prediction level from EEG signals using deep learning models

Kingphai, Kunjira and Moshfeghi, Yashar (2021) Mental workload prediction level from EEG signals using deep learning models. In: The 3rd Neuroergonomics Conference 2021, 2021-09-11 - 2021-09-16.

[thumbnail of Kingphai-Moshfeghi-NEC21-2021-Mental-workload-prediction-level-from-EEG-signals-using-deep-learning-models]
Preview
 Text. Filename: Kingphai-Moshfeghi-NEC21-2021-Mental-workload-prediction-level-from-EEG-signals-using-deep-learning-models.pdf
Accepted Author Manuscript
License: Strathprints license 1.0
Download (603kB)| Preview

Abstract

This paper describes our research groups’ efforts in tackling the mental workload (MWL) prediction task from Electroencephalogram (EEG) signals which was organised as a Passive BCI Hackathon grand challenge1 at Neuroergonomics conference 2021. This challenge focuses on the Multi-Attribute Task Battery-II (MATB-II), which has been used for assessing subject MWL capacities; since the MWL has been realised as an essential factor in subject performance within a complicated working system. However, decoding MWL levels (i.e. easy, medium and difficult) from EEG signals is a difficult task. Especially when the training and testing sets are recorded on different sessions, the distribution of the dataset could vary across them (Yin and Zhang, 2017). Thus, this competition was designed to develop algorithms to classify the three MWL levels from EEG signals in an unseen Session. This paper explores the potential of deep learning models to tackle this challenge. The deep learning models explored are Gated Recurrent Unit (GRU), Bidirectional GRU (BGRU), BGRU-GRU, Stacked Long-Short Term Memory (LSTM), Bidirectional LSTM (BLSTM) and BLSTM-LSTM.

ORCID iDs

Kingphai, Kunjira and Moshfeghi, Yashar ORCID logoORCID: https://orcid.org/0000-0003-4186-1088;
CORE (COnnecting REpositories)