Classification of EEG Signal for Body Earthing Application

Full text

e-ISSN: 2289-8131 Vol. 10 No. 1-2 81
Classification of EEG Signal for Body Earthing
Application
Noor Aisyah Ab Rahman, Mahfuzah Mustafa, Rosdiyana Samad, Nor Rul Hasma Abdullah,
Norizam Sulaiman and Dwi Pebrianti
Faculty of Electrical & Electronics Engineering, Universiti Malaysia Pahang, 26600 Pekan, Pahang, Malaysia
mahfuzah@ump.edu.my
Abstract—Stress is the way our body reacts to the threat and
any kind of demand. Stress happens when your nervous system
releases the stress hormones including adrenaline and cortisol
that lead to an emergency response of the body. Body earthing
technique is used to resolve this problem. Body earthing is a
method that is used to neutralize positive and negative charge in
the human body by connecting to the earth. EEG signals can be
used to verify the positive effect of body earthing. This project
focuses on the classification of EEG signals for body earthing
application. First, EEG signals from human brainwaves were
recorded by using Emotive EPOC Headset, before and after
body earthing for the 30 subjects. The alpha band and the Beta
band were filtered by using Band-pass filter ‘Butterworth’.
After filtering, the threshold of signal amplitude was set in the
range of -100 μV to 100 μV in order to remove the noise or
artifact. For feature extraction, Short-time Fourier Transform
(STFT) and Continuous Wavelet Transform (CWT) were used.
Lastly, the Artificial Neural Network (ANN) model is employed
to classify EEG signal taken from samples, before and after the
body earthing. A number of neurons chosen for this project are
55 with the mean square error 0.0023738. The result showed
that Alpha band signals before body earthing are low compared
to after body earthing. Whereas, for the Beta band signals, the
result before body earthing is high compared to after body
earthing. The increased signals of the Alpha band show that
subjects are in relax state, while the decreased of Beta band
signals shows the sample in stress state. These results imply for
both features of STFT and CWT. Based on the confusion matrix,
the result for the ANN classification yields 86.7% accuracy.
Index Terms—Body Earthing; Classification; CWT; EEG
signal; STFT;
I. INTRODUCTION
Based on previous studies, the body earthing or grounding
can reduce pain and fatigue. The free electrons will be
absorbed into the body when they are directly connected to
the earth. The function of body earthing is to balance the
positive charge produced in the body cell to maintain the
neutral state of the body.
Ghaly and Teplitz stated in their paper that the effect of
grounding or body earthing when sleeping could result in the
change of cortisol level in the human body. For two months,
12 subjects that had sleep dysfunction, pain, and stress slept
by using conductive mattress pads, and this method is the
same as grounded in the Earth. Further, to analyze the result
of cortisol changes, the level of diurnal cortisol secretion was
measured and circadian cortisol were profiled. Results from
the project indicate that the level of cortisol during sleep at
night was reduced while the synchronization of cortisol
hormone discharges was more in the arrangement with the
normal 24-hour circadian rhythm profile [1].
Chevaler et al. carried out a study to investigate the earthing
effects on human physiology. The conductive adhesive patch
was placed in the sole of 58 subjects’ feet. The parameters of
electrophysiological and the physiological were recorded
using biofeedback system. Body earthing was carried out for
28 minutes for each sample in two conditions: before and
after body earthing. After the earthing, half of the subjects
demonstrated an immediate change in RMS of estimations of
EEG signal from the left hemisphere, but not the right
hemisphere at all frequencies. Earthing the human body
indicated significant consequences for electrophysiological
properties of the cerebrum and musculature on the BVP of the
electrophysiological recordings. The progressions in EEG,
EMG, and BVP recommended decreases in the general level
of stress and a movement in ANS adjusted upon the body
earthing [2].
EEG is an electrophysiological checking technique to
record the electrical movement of the cerebrum. EEG is
utilized to gauge voltage changes from ionic current inside
the neurons of the brains [3]. EEG is additionally the
recording of cerebrum unconstrained electrical activity over
the time allotment. EEG can recognize the combinations of
numerous neurons that transmit signal, while making an
immense measure of electrical activity in the brains.
The limit of EEG is to record the neuronal activity in the
mind that prompts electric fields to the surface of the head
with a high temporal resolution. Along these lines,
electroencephalographic perusing is a non-intrusive
framework that can be associated with patients, ordinary
adults, and adolescents with no hazard or restrictions. EEG
signal, which is transmitted by human's cerebrum is a mixture
of activities, disposition, and different emotions.
Signal intensity of the EEG signal is quite small and
measured in microvolts (µV). The main frequency of human
EEG signal waves is delta, theta, alpha and beta [4]. Delta
waves have a recurrence of 3 Hz or underneath. It has a
tendency to be the most elevated inadequacy and the slowest
waves. It is ordinary as the predominant mood in babies up to
one year and in stages 3 and 4 for the rest. Next, theta has a
recurrence of 3.5 to 7.5 Hz and moderate action is delegated.
It is flawlessly typical in children up to 13 years and during
resting period, but an unusual occurrence for grown-ups
during their alert condition. Besides, alpha has a frequency
around 7.5 and 13 Hz, and generally the best found are is in
the back areas of the head on every side. Lastly, a normal
rhythm is during beta waves frequency at 14 Hz and above. It
is the dominant rhythm in patients who are alert or anxious or
have their eyes open.
During the research, it is important to choose a suitable
equipment for capturing the human brain waves signal and

Journal of Telecommunication, Electronic and Computer Engineering
82 e-ISSN: 2289-8131 Vol. 10 No. 1-2
transmitted it to the PC. Emotiv EPOC Headset is an EEG
sensor headset that reads the signals and a computer-based
system capable of processing the readings. The raw EEG data
can be collected by using Emotiv EPOC Headset. The placing
of several electrodes on the scalp is to measure the variety of
the electric potential created by neuronal action. Emotiv
EPOC Headset is also a low-cost brain PC interface for
human PC connection created by Emotiv Corporation. It is a
high-resolution device mounted on the head, which consists
of 16 electrodes, while 2 electrodes are used for reference and
have Bluetooth USB adapter [5].
EEG recordings were taken for both left and right
hemispheres at points Fp1 (AF3) and Fp2 (AF4) in the
international 10-20 system of electrode placement. The 10/20
system is an internationally recognized method to describe
the location of scalp electrodes. In this project, placement of
electrodes is based on 10/20 system. The number of ‘10’ and
‘20’ refers to the distances between the adjacent electrodes,
in which the distance are either 10% or 20% of the total front
back or right left from the skull. These electrodes are labelled
with letters, whereas the right hemisphere with even numbers
and left hemisphere with the odd numbers. A high power of
Alpha power shows the relax state of brain activity, while the
low power indicates the active state of brain activities.
A. Zabidi et al. mentioned that the technique to study the
characteristics of EEG signal in frequency and time domain
simultaneously is by using time-frequency analysis [6,7] The
most basic form of time-frequency analysis is STFT that
describes the imperative of frequency and spectral content of
signal at each point. Spectrogram was used to do the analysis
of the signal in time-frequency domain by applying the STFT
on the signal. Then, the signal is mapped into the 2D function
of time and frequency.
Based on the studies conducted by Tzallas et al., several
time-frequency analyses were used to classify EEG signal for
epileptic seizures. All of the methods will be compared to the
EEG signal analysis. STFT and 12 different time-frequency
distributions will be used to calculate the PSD of each
partition and the results were compared. STFT gives a good
result on the classification of the problem [8].
Research done by M.Akin is to investigate the wavelet
whether the transform is good for the spectral analysis.
Wavelet transform has been compared with FFT to see which
one is better as the spectral analysis tool of EEG signal. It was
found that the WT is more suitable to analyze the EEG signal
depends on the scaling and shifting of the mother wavelet [9].
M. Kemal Kiymik applied the STFT and CWT to the EEG
signals from normal person and child who are having an
epileptic seizure. Both outcomes were analyzed, and it was
resolved that the STFT was more relevant for the constant
handling of EEG signs due to its short procedure time. In any
case, the CWT still had a good resolution and the execution
is sufficiently high to be used in clinical and research settings
[10].
Based on a research done by Abdul Hamit Subasi and
Ergun Ercelebi, they used ANN and LR to classify the EEG
signal. In the study, two different classification models were
introduced. One of the methods is the conventional statistical
technique that used LR. Another technique is the rising
computationally capable systems based on ANN [11-13]. The
comparison between the created classifiers was essentially in
view of the investigation of the ROC curves and also various
scalar performance measures relating to the grouping. The
MLPNN based classifier was more accurate compared to the
LR based classifier [11].
Belakhdar et al. contemplated and assessed the
performances of two classifiers, which are the ANN and
SVM. The project goal is to dive deep into the examination
and to direct the suitable enhancement prompting an
expansion in the accuracy. In computing the components
vector, FFT was utilized. The vector contains 9 highlights, in
which these elements were then inputted to ANN and SVM
classifier to choose the most fitting one. ANN classifier was
the best compared to the SVM [14,15].
II. RESEARCH METHOD
First, EEG signals from human brainwaves were recorded
using Emotive EPOC Headset, before and after body
earthing. The alpha band and the Beta band were filtered by
using Band-pass filter ‘Butterworth’. After filtering, the
threshold of signal amplitude was set in the range of -100 μV
to 100 μV in order to remove the noise or artefact. For feature
extraction, STFT and CWT were used. Lastly, the ANN
model is employed to classify EEG signal taken from the
samples, before and after the body earthing process.
A. Data Collection
The 30 subjects were randomly chosen from the students in
Universiti Malaysia Pahang (UMP), who were between 19 to
23 years old. For the EEG signals recording for 6 minutes per
subject, students must be free from consuming any types of
medicine or drug. It is to make sure that the accuracy of the
data will not be affected. Subjects also not wear hair gel or
spray before collecting the data because the electrode cannot
attach to the scalp if the spray and gel were used. In addition,
subjects must be alert during the data collection session.
The raw EEG data can be collected using Emotiv EPOC
Headset. The purpose of placing of several electrodes on the
scalp is to measure the variety of the electric potential created
by neuronal action. Emotiv EPOC Headset is also a low-cost
brain PC interface for human PC connection created by
Emotiv Corporation. It is a high-resolution device that
mounted on the head, which consisted of 16 electrodes, while
2 electrodes were used for reference and have Bluetooth USB
adapter.
The 10/20 system is an internationally recognized method
to describe the location of the scalp electrodes. In this project,
placement of electrodes is based on 10/20 system. The
number of ‘10’ and ‘20’ refers to the distances between
adjacent electrodes, which are either 10% or 20% of the total
front back or right left from the distance of the skull. These
electrodes are labelled with a letter, whereas the right
hemisphere with even numbers and left hemisphere with odd
numbers. A high power of Alpha power shows the relax state
of brain activity, while the low power indicates the active
state of brain activities.
B. EEG Signal Pre-processing
All EEG data were pre-processed by using MATLAB
software. The Alpha and Beta band in EEG signals were
filtered using the Bandpass filter. The Band-pass filter
“Butterworth” is chosen because it has the flattest and there
is no ripple passband. The EEG raw data were filtered in the
frequency range of 8 Hz to 13 Hz in order to get the Alpha
band signals. However, the EEG raw data were filtered in the
frequency range of 13 Hz to 30 Hz to get the Beta band
signals, so the frequency outside the range will be filtered.

Classification of EEG Signal for Body Earthing Application
e-ISSN: 2289-8131 Vol. 10 No. 1-2 83
After filtering, the threshold of signal amplitude was set in
the range of -100 μV to 100 μV to remove the noise or artefact
that will affect the accuracy of the result.
C. EEG Signal Processing
In signal processing, the time-frequency based analysis was
used. Variety of the time-frequency analysis methods can be
used such as STFT, WV, IPS and CWT [16]. In this project,
STFT and CWT were used as the methods for the time-
frequency based analysis.
a. Short-Time Fourier Transform
STFT permits observing the changes in the frequency
spectra with reliance on the time. The sampled data are
partitioned into fragments by windowing function [17].
The time and the frequency resolution are specifically
identified with the window size. The shorter window
implies better time resolution yet the frequency resolution
will be worse. The time resolution can be expanded by
window overlapping.
b. Continuous Wavelet Transform
The CWT gives an option signal representation with an
adaptable time-frequency resolution. For the development
of non-stationary signals, the mathematical functions that
can be used are Wavelets [18]. The wavelet transform
performs a decomposition of the analyzed signal, x(t) into
a group that is localised in time and frequency.
D. Artificial Neural Network
ANN consists of a simple processing unit, which is neurons
or cells that are communicated by transmitting signals to each
other over a large number of weighted connections. Inputs for
the processing units are received from an external source or
neighboring unit and it is used to compute output signal to
send to other unit.
There are three types of units in neural network systems,
which are the input layers, hidden layers and output layers.
Two layers of Feed-Forward Neural Network were used in
this project. The network was trained using scaled conjugate
gradient backpropagation. The input vector is represented by
features arranged in 4 x 120 matrix. The information vector
is combined with a 2 x 120 target grid for training. The 120
sections were acquired from the number of sections
multiplied by the number of subjects. The components were
partitioned into 80% for training set, 5% for validation set and
15% for testing set. The analysis was performed using Neural
Network Toolbox in MATLAB. The ANN performances
were evaluated by means of MSE and Confusion Matrix.
III. RESULT AND DISCUSSION
Figure 1(a) and Figure 1(b) show the spectrogram
representation of the Alpha band and Beta band for AF3
respectively. Figure 1(c) and Figure 1(d) are the spectrogram
representation for AF4 channel. Based on Figure 1(a) to
Figure 1(d), the highest energy is represented by the red color.
The figure shows that the highest energy which is between
the frequencies of the band. For Alpha band, the signals are
between 8 Hz to 13 Hz, while for the Beta band the signals
are between 13 Hz and 30 Hz.
Figure 1(a): STFT spectrogram for Alpha Band in AF.
Figure 1(b): STFT spectrogram for Beta Band in AF3.
Figure 1(c): STFT spectrogram for Alpha Band Signal in AF4.
Figure 1(d): STFT spectrogram for Beta Band Signal in AF4.
Figure 2(a) and Figure 2(b) show the scalogram
representation of the Alpha band and Beta band for AF3
respectively. Figure 2(c) and Figure 2(d) are the scalogram
representation for AF4 channel. Based on Figure 2(a) to
Figure 2(d), the highest energy is represented by the red color.
The figure shows that the highest energy is between the
frequencies of the band. For Alpha band signals are between
8 Hz to 13 Hz, while for the Beta band is 13 Hz to 30 Hz. The
time range is from 100ms to 200ms.

Journal of Telecommunication, Electronic and Computer Engineering
84 e-ISSN: 2289-8131 Vol. 10 No. 1-2
Figure 2(a): CWT scalogram image for Alpha Band in AF3.
Figure 2(b): CWT scalogram image for Beta Band in AF3.
Figure 2(c): CWT scalogram image for Alpha Band in AF4.
Figure 2(d): CWT scalogram image for Beta Band in AF4.
Figure 3 shows the comparison of CWT and STFT based
on the energy extracted. The result shows the value of energy
extracted from the CWT is high compared to the STFT. This
is because CWT can extract the wavelet coefficients at certain
frequency, and this is very useful to monitor the critical
frequency components to the performance of the structure. In
terms of releasing the stress, CWT and STFT have shown the
same result.
Figure 4 shows the training performance and prediction
accuracy with varying numbers of neurons in hidden layers.
The variation number of neurons used between 5 and 150.
The number of neurons with the lowest MSE of AF4 channel
is chosen for this project. From ROC Curve in Figure 5, class
1 is the stress state, while class 2 is for the non-stress state.
Figure 3: Comparison between CWT and STFT in terms of Alpha-band and
Beta-band.
Figure 4: Training performance and prediction accuracy with varying no of
neurons in hidden layers.
Figure 5: Condition when wheel zigzag gets over block.
Based on Figure 6, the first two diagonal cells show the
number and correct percentage classifications by the trained
network. As showed in Figure 6, 13 subjects are correctly
classified as stress condition. This corresponds to 43.3% of
all the 30 subjects. As for the non-stress condition, 13
subjects are correctly classified as non-stress condition and
the percentage of correct classifications is 43.3%.
Three subjects are incorrectly classified as stress and it
corresponds to 10% of all the 30 subjects, while 1 subject
which is incorrectly classified as non-stress corresponds to
3.3% of all data. Out of 14 stress predictions, 81.3% are
correct and 18.8% are wrong and out of the 16 non-stress
predictions, 92.2% are correct and 7.1% are wrong.
Alpha Beta Alpha Beta
CWT STFT

Classification of EEG Signal for Body Earthing Application
e-ISSN: 2289-8131 Vol. 10 No. 1-2 85
From the 14 of the stress cases, 92.9% are correctly
predicted as stress and 7.1% are predicted as non-stress. Out
of 16 non-stress cases, 81.3% are correctly classified as non-
stress and 18.8% are classified as stress.
Based on the confusion matrix, the overall result of
classification yields 86.7% of predictions are correct and
13.3% are wrong classifications.
Figure 6: Condition when wheel zigzag gets over block.
IV. CONCLUSION
This project is to observe the effect of the body earthing to
the human brainwaves and emotions. The frequency band of
Alpha and Beta band were being analyzed before and after
the body earthing by implementing the MATLAB. The
objectives of this project are achieved through 4 stages that
are the pre-processing of EEG signals, analysis of EEG signal
by using Short Time Fourier Transform, Continuous Wavelet
Transform and classification using Artificial Neural Network.
The result shows that the objectives, which are to compare
the time-frequency analysis of STFT and CWT for EEG
signal in body earthing applications and to classify EEG
signal for body earthing applications are achieved. Based on
the result, before body earthing the Alpha band signal is high
and the Beta band signal is low. Alpha band signals are
gradually increased and Beta band signals are decreased. The
increased Alpha band signal activities show that the subjects
feel more relaxed. The decreased of Beta band signal
activities show that subjects are in alert and stress state.
For the recommendation, the different technique for EEG
signals analysis. There are many types of time-frequency
based analysis that can be used. For the classification of EEG
signal, another intelligent classification can be used, such as
CBR and fuzzy method. In data acquisition, the protocol for
EEG signal data acquisition must be followed in order to
obtain the accurate result. This is because if all the protocol
followed the noise or artefact in the signal, it can be reduced.
ACKNOWLEDGMENT
This research was supported by Universiti Malaysia
Pahang (RDU170365).
REFERENCES
[1] M. Ghaly and D. Teplitz, “The biologic effects of grounding the human
body during sleep as measured by cortisol levels and subjective
reporting of sleep, pain, and stress,” J. Altern. Complement. Med. vol.
10, pp. 767–776, 2004.
[2] G. Chevalier, K. Mori, and J. L. Oschman, “The effect of earthing
(grounding) on human physiology,” Eur. Biol. Bioelectromagn. vol.3,
pp. 600–621, 2006.
[3] P. M. Sovilj, S. S. Milovancev, and V. Vujicic, “Digital Stochastic
Measurement of a Nonstationary Signal With an Example of EEG
Signal Measurement” IEEE Trans. On Instrumentation and
Measurement, vol. 60, pp. 3230–3232, 2011.
[4] M. Mustafa, S. K. Huong, N. Sulaiman, R. Samad, N. R. H. Abdullah,
N. S. Pakheri and M. N Taib, “Initial result of body earthing on student
stress performance,” in Int. Conf. BioSignal Anal. Process. Syst.
ICBAPS 2015, pp.129–133.
[5] D. S. Benitez, S. Toscano, and A. Silva, “On the use of the Emotiv
EPOC neuroheadset as a low cost alternative for EEG signal
acquisition,” in 2016 IEEE Colomb. Conf. Commun. Comput.
COLCOM, pp. 1-6.
[6] A. Zabidi, W. Mansor, Y. K. Lee, and C. W. N. F. C. W. Fadzal, “Short-
time Fourier Transform analysis of EEG signal generated during
imagined writing,” in Syst. Eng. Technol. (ICSET) Int. Conf. no. 2. pp.
1–4.
[7] M. Snajdarova, B. Babusiak, and I. Cap, “Interactive Tool for Analysis
of EEG Signal,” ELEKTRO. pp.583-588, 2016.
[8] A. T. Tzallas, M. G. Tsipouras, and D. I. Fotiadis, “Epileptic seizure
detection in EEGs using time-frequency analysis,” IEEE Trans. Inf.
Technol. Biomed. vol. 13, pp. 703–710, 2009.
[9] M. Akin,”Comparison of wavelet transform and FFT methods in the
analysis of EEG signals,” J. Med. Syst. vol. 26, pp. 241–247, 2002.
[10] M. K. Kiymik, I. Güler, A. Dizibüyük, and M. Akin, “ Comparison of
STFT and wavelet transform methods in determining epileptic seizure
activity in EEG signals for real-time application,” Comput. Biol. Med.
vol. 35, pp. 603–616, 2005.
[11] A. Subasi and E. Erçelebi , “Classification of EEG signals using neural
network and logistic regression,” Comput. Methods Programs Biomed.
vol. 78, pp. 87–99, 2005.
[12] A. T. Tzallas, M. G. Tsipouras, and D. I. Fotiadis, “Automatic seizure
detection based on time-frequency analysis and artificial neural
networks,” Comput. Intell. Neurosci. PMC2246039, 2007.
[13] R. F. Navea and E. Dadios, “Classification of wavelet-denoised
musical tone stimulated EEG signals using artificial neural networks,”
in 2017 IEEE Reg. 10 Annu. Int. Conf. Proceedings/TENCON, pp.
1503–1508.
[14] I. Belakhdar, W. Kaaniche, R. Djmel, and B. Ouni, “A comparison
between ANN and SVM classifier for drowsiness detection based on
single EEG channel,” in 2nd Int. Conf. Adv. Technol. Signal Image
Process. ATSIP 2016, pp. 443–446.
[15] D. Garrett, D. A. Peterson, C. W. Anderson, and M. H. Thaut. 2003,
“Comparison of linear, nonlinear, and feature selection methods for
EEG signal classification,” IEEE Trans. Neural Syst. Rehabil. Eng. vol.
11, pp. 141–144, 2003.
[16] I. Yesilyurt, “The application of the conditional moments analysis to
gearbox fault detection - A comparative study using the spectrogram
and scalogram,” NDT E Int. vol. 37, pp. 309–320, 2004.
[17] M. Mustafa, M. N. Taib, Z. H. Murat, N. Sulaiman, and S. A. M. Aris,
“Classification of EEG spectrogram image with ANN approach for
brainwave balancing application,” Int. J. Simul. Syst. Sci. Technol. vol.
12, pp. 29–36, 2011.
[18] N. Hazarika, J. Z. Chen, A. C. Tsoi and A. Sergejew, “Classification of
EEG signals using the wavelet transform,” Signal Processing. vol. 59,
pp. 61-72,