An Improved InceptionTime Model for Mental Workload Assessment Based on EEG Signal

The assessment of mental workload is pivotal technology in the field of human-machine interaction. In this paper, we propose a novel convolutional neural network (CNN) architecture designed for EEG-based mental workload classification. The backbone of our proposed model is built upon the InceptionTime network, a cutting-edge network specifically tailored for time-series classification. Then, a squeeze-and-excitation network is added to the model to automatically focus on channels that are more useful for mental workload classification task. Furthermore, we conduct ablation experiments to evaluate the impact of different model components, showcasing their individual effectiveness. On a public dataset of EEG signals for mental workload, the proposed model can achieve an accuracy of 70.89% and an Fl score of 72.61 %. In the ablation experiment, we observed a significant decrease in the model's performance upon removing the convolutional layer with a kernel size of 9 and this highlights the crucial role of high-frequency signal information extraction in the model for the mental load classification task. In addition, by fine-tuning the model with varying dataset sizes, we demonstrate the model's robust classification capability even with limited data. In conclusion, our improved InceptionTime model demonstrates significant potential as a proficient classifier for EEG-based mental workload tasks.