IEEE Systems, Man and Cybernetics Magazine - January 2022 - 5

different fields including education and training [1], [2];
military applications [2], [3]; performance assessment in
competitive sports [4], [5]; aged care [6], [7]; medical diagnostic
processes [8]-[11]; human machine collaboration
[12]; and robot teleoperation [13]. In [14], Salaken et al.
used four different classifiers: random forests (RFs), neural
networks (NNs), linear discriminant analysis (LDA),
and logistic regression for cognitive load detection. In
their research, they have used EMOTIV EPOC+ as an electroencephalogram
(EEG) device and EmotivPro software
for automatic power spectral features calculation.
They have achieved 95% accuracy in real time at 8 Hz frequency
for the RF classifier. The multiclass EEG data
from the brain-computer interface application is analyzed
in [15], where they combine two stages, the common
spatial pattern is used for feature extraction, and
fuzzy logic is used for classification, which showed excellent
accuracy. Fridman et al. [16] proposed two vision-based
methods to automatically estimate the cognitive load nonintrusively
under real-world conditions with a hidden Markov
model (HMM) and 3D-convolutional NNs (CNNs). On a
data set of 92 subjects, the 3D-CNN approach outperforms
the HMM approach with 86.1% accuracy. Krejtz et al. [17]
tested the sensitivity and reliability of pupil diameter and
microsaccades as the indicator of the cognitive load and
demonstrated that they can be implemented to discriminate
among different levels of cognitive load. Armougum
et al. [18] simulated a virtual reality model of a train station
to compare the cognitive load during navigation
between virtual and real situation. The measurement was
done on physiological, subjective, and behavioral aspects
of cognitive load using electrodermal activity, NASA-Task
Load Index, and recognition of relevant factual and contextual
information seen by travelers in the train station,
respectively. It was conclusive that virtual reality provided
a great opportunity to measure cognitive load during
navigation.
Researchers in [19] compared Riemannian geometrybased
classifiers and CNNs for cognitive or affective states
classification. New variants of algorithms were proposed
and benchmarked with classic methods to estimate both
cognitive load and affective states (valence/arousal) from
EEG signals with subject-specific and subject-independent
calibration to determine the possibility of calibration-free
systems. A guideline to the usage of different methods in
different situations was also introduced in this study. To
lessen the difficulty of the time-consuming setup process
of wet electrodes and poor data quality due to artifacts
originating from noncerebral origins, researchers in [20]
designed and implemented several types of novel dry-contact
EEG sensors that can efficiently reduce preparation
time without the necessity for any conductive gel. The new
electrodes allow users to move around freely in any operational
environment. Thomas and Vinod [21] highlighted
up-to-date achievements in the fields of EEG-based biometric
systems. Also, they tried to figure out the future
directions and challenges in applying the technology in
practical situations.
With the current progress of wearable technologies ,
EEG sensing devices, Internet of Things, and machine
learning embedded systems, it can easily be extrapolated
that cognitive load identification process will be utilized
widely across different areas in the industry and the process
itself will be more portable and simpler over the
upcoming years. These assumptions are based on the fact
that, recently, the introduction of wireless saline-based
headsets that take only a few minutes of setup time has
made it much easier to collect EEG data from human
brains. We take advantage of brain signals, one of the most
efficient indicators of cognitive functions, in this work.
Over the past few years, many methods have been
established to classify EEG, including hidden Markov
models [22], fuzzy systems [23], optimal spatial filters [24],
wavelet packet feature extraction [25], NNs [26], genetic
January 2022 IEEE SYSTEMS, MAN, & CYBERNETICS MAGAZINE
5
IEEE Systems, Man and Cybernetics Magazine - January 2022

Table of Contents for the Digital Edition of IEEE Systems, Man and Cybernetics Magazine - January 2022
