EEG-based mental workload recognition using deep learning techniques

In most cases, mental workload (MWL) refers to the cost of cognitive resources in a certain task. [1]. A high MWL means the subject uses most of cognitive resources on the given task. Understanding the MWL level induced is important for optimizing human resources and reducing accidents. But in most cases, it is hard to measure the MWL accurately. EEG signal has a long history in medical diagnose. During the brain activities, electric potential generated by large number of neurons synchronizes together and forms EEG. EEG is a good indicator of cortex activities. In the past a few years, small wearable EEG devices became available in the market. These devices made it convenient to record the EEG. Recent years, encouraged by the convenience of EEG, there were many studies on EEG based mental workload analysis. This paper reviews and compares common EEG based MWL recognition algorithms in the past 3 years (2018 to 2020). A review of the existing public EEG datasets is also proposed. Then, this paper uses STEW dataset [2] to test some common algorithms. The result shows that for classification between resting and under SIMKAP task [3] of the same subject (subject dependent classification), many existing algorithms can achieve good result. For more detailed classification (low, medium, high, 3 classes) of different subjects (subject independent classification), the best result achieved by this paper is only about 65%. Although due to many reasons, the results achieved by this paper were not good enough, the review of the state of art algorithm in chapter 3 shows EEG based MWL recognition has big potential and not far from commercial application.