Sha’abani, Nurul Al Hafiz (2022) Mental workload classification on two frontal eeg channels using artificial neural network with peak frequency based intrinsic mode function selection. Doctoral thesis, Universiti Tun Hussein Onn Malaysia.

Preview	Text 24p MOHD NURUL AL HAFIZ SHA’ABANI.pdf Download (549kB) | Preview
	Text (Copyright Declaration) MOHD NURUL AL HAFIZ SHA’ABANI COPYRIGHT DECLARATION.pdf Restricted to Repository staff only Download (369kB) | Request a copy
	Text (Full Text) MOHD NURUL AL HAFIZ SHA’ABANI WATERMARK.pdf Restricted to Registered users only Download (33MB) | Request a copy

Abstract

Recently, the cognitive neuroscience research community has been interested in measuring mental workload based on electroencephalogram (EEG) signals. In multichannel EEG studies, it is reported that the signal from the brain's frontal area is associated with mental workload. Hence, using a single or a few EEG channels could be feasible. However, the challenge is to remove the presence of artefact in the EEG signal since not all existing methods in multichannel EEG work well when using a single or a small number of channels due to the limited signal sources. To overcome the limitation, this thesis proposed using the ensemble empirical mode decomposition (EEMD) technique. The EEMD heuristically decomposed the EEG signal in the time domain into a series of intrinsic mode functions (IMF). Here, a new technique is proposed for selecting the significant IMFs based on the peak frequency of power spectral density. To evaluate the performance of the proposed method, a mental arithmetic task and a building structure task are designed to stimulate the mental workload of cognitive and psychomotor activities. A cross-task and a cross-subject classification based on the designed mental workload task have been done. Relative power, Shannon entropy, log energy entropy, skewness and kurtosis were extracted as features. Results show that the EEMD coupled with ANN provides the best overall performance compared to the other classifiers. In cross-task classification, the classification accuracy achieved 90.0% when using the log energy entropy feature. While in cross-subject classification, the accuracy achieved 76.7% and 86.7% for mental arithmetic and building structure tasks, respectively when the Shannon entropy feature was used. In a comparative study, the EEMDANN and log energy entropy features provide better accuracy in cross-task and cross-subject classification than the previous studies. In conclusion, this work has contributed to the EEG-based mental workload evaluation using a small number of EEG channels

Actions (login required)

View Item