IEEE Consumer Electronics Magazine - September/October 2020 - 37

normal state. In the context of real time stress
monitoring, there are several design challenges
associated with traditional cloud computing like
latency, energy consumption, cost, security, etc.
To address these design challenges, several
works have used edge computing as a potential
solution.

Figure 3. Common features extracted from ECG,
GSR, Respiration and EEG.

Figure 4 shows an overview of a typical stress
classification framework. The signals from the
wearable sensors are preprocessed and subsequently features are extracted from the processed signals that are used to train a machine
learning model to classify between stressed and

EDGE COMPUTING FOR STRESS
MONITORING
The idea of edge computing is to migrate
some of the computing capacity close to the endpoint where data is collected also known as the
edge. When computations are done at the edge,
the response time from the system reduces considerably as the data does not need to transmit
through a very long distance. Also, techniques

Table 3. Related work and attributes.
Year

Subjects

Signals

ML Algorithm

Features

AW(s)

Classes

Accuracy(%)

Palanisamy et al.15

2013

10

ECG

k-NN

12

32

2

94.58

Muaremi et al.16

2014

10

ECG, Resp, ST, EDA, Activity

SVM, k-NN, ANN, RF

187

NA

2

73

Liapis et al.17

2015

31

EDA

LDA

21

NA

2

98.8

2015

12

EDA

LR

NA

NA

2

91.66

2015

5

BVP, HRV, EDA

SVM

NA

0.1

2

80

2015

7

EDA, EMG, HR, Resp

SVM

16

200

3

98.41

2015

67

ECG, Resp, Activity

SVM

22

NA

2

72

2016

10

ECG

k-NN

12

32

2

94.58

2016

50

HR, Resp, EDA, ST, TF

DT

59

NA

2

89.07

Lee et al.24

2016

8

Activity

SVM

14

300

2

83.34

Gjoreski et al.25

2016

5

BVP, ST, EDA, HR

RF

63

240

2

76

Mozos et al.26

2017

18

EDA, BVP, Speech, Activity

AdaBoost

NA

NA

2

94

2017

7

EDA, BVP

SVM, LR, RF

110

60

2

88.8

2017

28

BVP, Activity

SVM

20

NA

2

95

2017

9

ECG, EDA, Resp

SVM

NA

100

3

99

Zubair et al.

18

Sandulescu et al.
Ghaderi et al.

19

20

Hovsepian et al.
Castaldo et al.

21

22

Abouelenien et al.

Egilmez et al.
Lee et al.

23

27

28

Chen et al.
Betti et al.

29

30

2018

12

HRV, EDA, EEG

SVM

15

300

2

86

Rachakonda et al.

31

2019

NA

Image Inputs

CNN

4

NA

2

97

Rachakonda et al.

32

2019

NA

EDA, Activity, ST

DNN

NA

NA

2

98.3

2020

13

EDA, PPG

RF

5

30

2

92

Nath et al.33

EDA=Electrodermal Activity, EMG=Electromyogram, Resp=Respiration, HR=Heart Rate, ECG=Electrocardiogram, BVP=Blood Volume Pulse,
HRV=Heart Rate Variability, ST=Skin Temperature, PD=Pupil Diameter, LDA=Linear Discriminant Analysis, SVM=Support Vector Machine,
FDA=Fisher Discriminant Analysis, k-NN=k-Nearest Neighbor, LR=Logistic Regression, DT=Decision Tree, ANN=Artificial Neural Network,
RF=Random Forest, CNN=Convolutional Neural Network, AW=Acquisition Window, ML=Machine Learning, NA=Not Available.

September/October 2020

37
IEEE Consumer Electronics Magazine - September/October 2020

Table of Contents for the Digital Edition of IEEE Consumer Electronics Magazine - September/October 2020
