579
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
IDENTIFICATION OF DRIVERS DROWSINESS BASED ON
FEATURES EXTRACTED FROM EEG SIGNAL USING SVM
CLASSIFIER
M.Thilagaraj
Department of Electronics and Instrumentation Engineering Karpagam College of Engineering
Coimbatore, Tamilnadu, (India).
E-mail: m.thilagaraj@gmail.com
ORCID: https://orcid.org/0000-0002-7729-3273
M. Pallikonda Rajasekaran
Department of Electronics and Communication Engineering, Kalasalingam Academy of Research and
Education Krishnankoil, Tamilnadu, (India).
E-mail: m.p.raja@klu.ac.in
ORCID: https://orcid.org/0000-0001-6492-4512
U. Ramani
Department of Electrical and Electronics Engineering, K. Ramakrishnan College of Engineering
Trichy, Tamilnadu, (India).
E-mail: ramani.eee@krce.ac.in
ORCID: https://orcid.org/0000-0003-3820-3607
Recepción:
28/11/2019
Aceptación:
16/09/2020
Publicación:
30/11/2021
Citación sugerida:
Thilagaraj, M., Rajasekaran, M. P., y Ramani, U. (2021). Identication of drivers drowsinessbased
on features extracted from EEG signal using SVM classier. 3C Tecnología. Glosas de innovación aplicadas
a la pyme, Edición Especial, (noviembre, 2021), 579-595. https://doi.org/10.17993/3ctecno.2021.
specialissue8.579-595
580
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
ABSTRACT
Electroencephalogram (EEG) is a recording machine used for storing the electrical
movement of the brain. The brain waves are produced by passing electric current through
the brain and that is being recorded by the Electroencephalogram. After taking the EEG
signal the process of removing the noise and low-quality signal is carried out by using the
Butterworth lter and that process is known as Preprocessing. Then the signal is segmented
with the help of Discrete Wavelet Transform (DWT) so that the signals are segmented into
ve primary frequency bands (delta, theta, alpha, beta, and gamma). Finally, the EEG
signals were classied based on the statistical features obtained from the dierent segments
of the EEG signals using Support Vector Machine Classier. SVM maps input vector to
a high dimensional space where a nest hyper plane is developed. Among the numerous
hyper planes accessible, there is only one hyper plane that amplies the separation among
itself and the closest information vectors of every class. The identication of the fatigue
based on the features extracted using SVM is more ecient compared to the other feature
extraction methods employed for the analysis of the signals.
KEYWORDS
ElectroEncephaloGram (EEG), Discrete Wavelet Transform (DWT), Statistical Features,
SVM Classier.
581
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
1. INTRODUCTION
Drowsiness is the intermediate state among awaken and sleep. Many factors can cause
tiredness or fatigue in driving as well as long driving hours, lack of sleep, take more
medicines, eating of alcohol, and some early morning drive, mid-afternoon hours, driving
at midnight, and particularly in a monotonous driving environmental setup. Driving under
the inuence of drowsiness will cause: Reduction in the level of concentration and reduction
in the ability of a person to take decision quickly.
According to the Nationwide Road travel protection management (NHTPM), about
1,00,000 crashes are the directly outcome of driver sleepiness each year. This is the basis
why more and more researches are going on in this eld. So to avoid such accidents this
project has been established.
The drowsiness had been diagnosed on the premise of (Electroencephalogram) EEG,
(Electrooculography) EOG, Galvanic Skin Response (GSR), coronary heart charge and
pulse price and so forth, these are some of the physiological measurements
For identifying drowsiness many strategies have been used typically for the process of
segmentation, tiny Time Fourier Transform (STFT) is used. Wigner–Ville Distribution
(WVD) is employed to extract features from the EEG indicators. The statistical features set
is being decomposed into small and nite range of intrinsic mode function by means of
empirical mode decomposition technique. In the existing works benchmark datasets is used.
Characteristic extraction and function class are the two main modules of biomedical signal.
For studying non stationary signals time frequency illustration is used. TFRS are spoken to
by method for either adequacy or vitality thickness throughout the years and recurrence.
This paper gives a brand-new approach primarily based at the aggregate of time-frequency
picture and DWT the SVM to categorize the EEG signal for fatigue detection. The EEG
indicators are to start with training and labeling the indicators by means of extracting
features from the segmented components. Then EEG signals are segmented via employing
DWT. The statistical features had been extracted from the segments. The evaluation
parameters include variance, skewness, entropy and kurtosis of the segmented signals.
The extracted capabilities have been after that classied using SVM Classier. The overall
582
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
presentation of the process is estimated through guring exactness, and explicitness of the
classier. Right here each nonlinear and non-stationary signal are evaluated by using DWT.
As a method to break down markers is posteriori and grants the mining of the inward sizes
of each sign, so it has a rst-rate benet in EEG signal processing. It aords an automatic
popularity of fatigue based at the time-frequency image the usage of DWT and SVM.
The best capabilities of alpha, beta, gamma, theta, and delta waves are fed into the sample
recognition tool for category of fatigue and non-fatigue EEG indicators.
2. MATERIALS AND METHODS
2.1. RELATED WORKS
Chen et al. (2015) proposed regular detection of awareness or sluggishness from
physiological sign utilizing wavelet based nonlinear highlights and articial intelligence
physiological indicator such as electroencephalogram and electrooculography recording is
much necessary non-invasive procedures of detecting someone’s awareness/tiredness. We
recommend a gadget for drowsiness detection the usage of physiological signals that gift
4 benets: (1) disintegrating EEG signals into wavelet sub-groups to extract extra glaring
information past uncooked signal (Chen et al., 2015), (2) extraction and combination of
nonlinear features from EEG sub-bands, (3) combination the facts from EEG’s and eyelid
actions, (4) utilizing procient very learning machine for notoriety order (Zhang, Wang, &
Fu, 2014). The preliminary outcomes demonstrate that the proposed system accomplishes
not handiest a high recognition exactness but rather additionally a completely quick
calculation speed (Bajaj & Pachori, 2012).
Zhang et al. (2014) proposed automatic nding of Driver tiredness Based on Entropy and
Complexity Measures. In this research work oers an actual-time approach based totally
on numerous entropy and intricacy measures for recognition and identication of using
exhaustion from recorded biological signal indicators (Kar, Bhagat, & Routray, 2010; Picot,
Charbonnier, & Caplier, 2012). The entropy-based totally capabilities, particularly, the
Wavelet Entropy (WE), the peak-to-top fee of apen (pp-apen), and the peak-to-height cost
of pattern entropy (pp-sampen), were extricated from the collected alarms to assess the
driving exhaustion levels (Kumar, Raju, & Kumar, 2012; Picot, Charbonnier, & Caplier,
583
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
2011). The actual-time capabilities received via we, pp-apen, and pp-sampan with sliding
window have been applied to synthetic neural community for education and trying out the
machine, which gives 4 conditions for the fatigue stage of the patients, specically, normal
country, mellow weakness, emotional episode, and over the top weariness. Then, the motive
force fatigue stage may be determined in actual time (Majumdar, 2011; Mardi, Ashtiani, &
Mikaili, 2011).
2.2. METHODOLOGY
This process includes ve important modules each of them have a unique processing
capability.
Figure1. Block diagram.
Source: own elaboration.
584
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
The ow diagram of the module is mentioned below as follows:
Figure 2. Flow diagram of modules.
Source: own elaboration.
2.3. PREPROCESSING
Preprocessing is carried out in EEG signal, why because EEG signals are not constant in
nature (i.e) dierent frequency workings are existed in dierent interval of time. The input
signal needs to be preprocessed before going to process the signal. To remove extracting
time, unwanted noise, frequency and TF/TS domain features from the multi-channel
EEG the process of preprocessing is carried out, with this technique systole, noise, low
quality signal will be removed. EEG signals are preprocessed using Butterworth lter. Here
Butterworth is used as both low pass and high pass lter. Before preprocessing DC, Drift
Elimination is carried out for removing drift.
Butterworth lter: Butterworth lter is an ideal lter. The denition of ideal lter is that
the lter that not only completely reject the unwanted frequencies but should maintain
a uniform sensitivity over the wanted frequencies such a lter cannot be obtained but
butterwort lter showed that by increasing the number of lter elements. Butterworth
585
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
lter can able to adjust the component values of the lter. That his basic low pass lter
functionality could be modied to give high pass, band stop, band pass functionality.
Butterworth lter has maximally at frequency response (i.e.) in the pass band it has no
ripple and in the stop band it rolls o towards zero. Compared to Chebyshev and Elliptic
lter, more linear phase response is achieved by the butter worth lter in the pass band.
586
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 3. (a) Input Signal, (b) DC Drift Elimination, (c) Preprocessed signal using HPF, (d) Preprocessed signal
using LPF.
Source: own elaboration.
2.4. SIGNAL SEGMENTATION
On the basis of frequency-bands of the rhythms Segmentation has been employed. EEG
signal can be taken into consideration as a superposition of various structure occurring
587
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
on distinct time scales at dierent time. There are two types of segmentation, the signal
segments to equal part are the rst type and is called constant segmentation. The advantage
of this method is that it is very simple to process, the disadvantage is bad accuracy. In the 2d
method for segmentation of non-stationary signal is adaptive segmentation which means
the signal segments robotically to wise components with equal residences.
Segmentation procedure segments the signal into ve primary frequency band they're delta,
theta, alpha, beta, gamma. The range of every frequency are delta ranges from 0.5Hz to
4Hz, theta ranges from 4Hz to 8Hz, alpha ranges from 8Hz to 12Hz and beta ranges from
12Hz to 30Hz Gamma ranges from greater than 30Hz
• Here Alpha, Beta sub bands are responsible for drowsiness.
• When a person is closing his eye or resting means, predominant of alpha activity
is carried out.
• Transition from unsleeping to sleep nation alpha waves decreases at the same time
as theta waves increases.
• Beta waves are excessive whilst someone is taking capsules.
The process of Segmentation is carried out using Discrete Wavelet Transform (DWT).
Discrete Wavelet Transform: Intrinsic multi-goals nature, wavelet-coding plans for
application where versatility and acceptable deprivation are important this is why we have
chosen separate Wavelet Transform. Wavelet change deteriorates a sign into a lot of premise
capacities. These premise capacities are called wavelets. Separate Wavelet Transforms
(SWT), which transforms a discrete time signal to a discrete wavelet symbol.
Basically, the medical images need more accuracy without loss of sequence. The Separate
Wavelet Transform (SWT) was based on time-scale representation, which provides capable
multi-resolution.
It has been analyzed that the Separate Wavelet Transform (SWT) operates at a maximum
clock frequency of 99.197 MHz respectively.
588
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 4. Discrete Wavelet Transform.
Source: own elaboration.
DWT oers end result a lot sharper than any of the conventional analysis method in the
time-frequency domain.
589
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Figure 5. (e) Step input (f) Signal segmentation.
Source: own elaboration.
2.5. FEATURE EXTRACTION
Feature Extraction is obtained from the segmented parts of the EEG signal by the process
of Signal segmentation. After completion of noise removal process in EEG signal, need to
extract the feature parameters from the EEG signal. Statistical features are extracted they
are Mean, Standard Deviation, Entropy, Skewness, Kurtosis.
Mean: In possibility, mean and expected value are used synonymously to consult one
measure of the vital tendency either of a possibility distribution or of the random variable
characterized via that distribution.
Standard Deviation: The Standard Deviation (SD) measures the amount of version or
dispersion from the average. a low standard deviation indicates that the information points
have a tendency to be very near the suggest (additionally known as expected fee); a high
standard deviation suggests that the facts factors are unfold out over a big variety of values.
Entropy: Entropy is the common quantity or amount of information of statistics contained
in each message acquired right here, message stands for an occasion, sample or person
drawn from a distribution or information circulation entropy consequently characterizes
our uncertainty about our source of records.
590
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Skewness: Skewness is a measure of the asymmetry of the opportunity of an actual-
valued random variables approximately it suggests.
Kurtosis: Relative to a normal sharing Kurtosis is a measure of whether the data are
pointed or level. If it has a particular top close to the average then it has data sets with high
kurtosis, decline rather rapidly, if it tends to have a at top near the mean rather than a
sharp max out then the data sets have a low kurtosis. A uniform distribution would be the
acute case.
Figure 6. Feature Extraction.
Source: own elaboration.
2.6. CLASSSIFICATION
Based on the features extracted from the signals the classication process is carried out
using SVM Classier. The Support Vector Machine is an administered AI calculation that
can be utilized to analyze information and perceive examples utilized for the procedure of
characterization. The basic SVM takes a set of data as input and for each given input the
SVM predicts, the two classes that forms the hyper plane, making it a non-probabilistic
double straight classier. For SVM set of training examples are carried out by marking that
belongs to one or two categories, so that it builds a model for classication.
591
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
SVM Classier:
Figure 7. SVM algorithm.
Source: own elaboration.
SVM maps input vectors to a higher dimensional vector space where an ideal overexcited
plane is built. Among the numerous hyper planes accessible, there is only one hyper plane
that expands the separation among itself and the closest information vectors of every
classication. This overexcited plane which amplies the edge is known as the most ideal
isolating overexcited plane and the edge is characterized as the total of separations of the
hyper plane to the nearest preparing vectors of every class. For identifying the right hyper
plane there are ve types of situations.
Expression for hyper plane
w.x+b = 0
x – Set of guidance vectors
w – Vectors perpendicular to the separating hyper plane
b – Oset parameter which allows the increase of the margin
The Support Vector Machine has the following advantages
• For clear margin of separation, the SVM really works well.
• For high dimensional spaces SVM is more eective.
Input Space
Features Space
Input Space
592
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
• In the event that the quantity of measurements is more noteworthy than the
quantity of tests the SVM is more eective.
• It is memory productive why since it utilizes a subset of preparing focuses in the
choice capacity called bolster vectors.
Figure 8. Classication.
Source: own elaboration.
3. RESULTS
The performance analysis of the Support Vector Machine is analyzed, compared to the
Neural Network Classier the accuracy of the drowsiness identication is increased. This
observation makes use of kernel capabilities which includes the gaussian radial foundation
characteristic and the polynomial feature. SVM makes the process more simple and more
ecient compared to neural network classier.
4. CONCLUSIONS
From the research work that has been carried out with the machine learning technique we
can identify that the best classication accuracy is achieved by the SVM classier. Wigner–
Ville distribution (WVD) is employed to extract features from the EEG signals. The EMD
technique decomposes a dataset into a nite and often small number of intrinsic mode
functions (IMFs) were employed to segment the EEG signals. The benchmarked dataset that
has been used in the existing works. A robust and computationally low-intensive wavelet-
based features have been used for the proposed EEG classication system.
The main advantage of the Support Vector Machine compared to other classiers is that
if you are working with an unbalanced data set then SVM is a good choice. In SVM the
computational complexity is reduced and classication eciency is increased compared to
other classiers mainly due to the global optimization functions included. The classier has
593
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
shown a terric live performance in class. DWT presents an awesome nearby portrayal of
the wavering added substances of the non-stationary or nonlinear sign. This system gives
preferable sort exactness over a couple of methodologies contemplated beforehand.
ACKNOWLEDGEMENT
This work was supported by the department of Electronics and Instrumentation department
of Kalasalingam Academy of Research and Education.
REFERENCES
Bajaj, V., & Pachori, R. B. (2012) Separation of Rhythms of EEG Signals Based on
Hilbert-Huang Transformation with Application to Seizure Detection. In: Lee,
G., Howard, D., Kang, J. J., & Ślęzak, D. (eds.), Convergence and Hybrid Information
Technology. ICHIT 2012. Lecture Notes in Computer Science, vol. 7425. Springer,
Berlin, Heidelberg. https://doi.org/10.1007/978-3-642-32645-5_62
Chen, L.-I., Zhao, Y., Zhang, J., & Zou, J.-Z. (2015). Automatic detection of alertness/
drowsiness from physiological signals using wavelet-based nonlinear features and
machine learning. Expert Systems with Applications: An International Journal, 42(21), 7344-
7355. https://dl.acm.org/doi/10.1016/j.eswa.2015.05.028
Gurudath, N., & Riley, H. B. (2014). Drowsy Driving Detection by EEG Analysis Using
Wavelet Transform and K-Means Clustering. Procedia Computer Science, 34, 400-409.
https://doi.org/10.1016/j.procs.2014.07.045
Kar, S., Bhagat, M., & Routray, A. (2010). EEG signal analysis for the assessment and
quantication of driver’s fatigue. Transportation Research Part F: Trac Psychology and
Behaviour, 13(5), 297-306, https://doi.org/10.1016/j.trf.2010.06.006
Kumar, G. R., Raju, S. V. P., & Kumar, D. S. (2012). Classication of EEG signals for
drowsiness Detection in brain and computer interface. GESJ: Computer Science and
Telecommunications, 4(36). https://www.researchgate.net/publication/261100256_
Classification_of_EEG_Signals_for_Drowsiness_Detection_in_Brain_and_
Computer_Interface
594
https://doi.org/10.17993/3ctecno.2021.specialissue8.579-595
3C Tecnología. Glosas de innovación aplicadas a la pyme. ISSN: 2254 – 4143 Edición Especial Special Issue
Noviembre 2021
Majumdar, K. (2011). Human scalp EEG processing: Various soft computing approaches.
Applied Soft Computing, 11(8), 4433-4447. https://doi.org/10.1016/j.asoc.2011.07.004
Mardi, Z., Ashtiani, S. N. M., & Mikaili, M. (2011). EEG-Based Drowsiness Detection
for Safe Driving Using Chaotic Features and Statistical Tests. Journal of Medical Signals
& Sensors, 1(2), 130-137. https://pubmed.ncbi.nlm.nih.gov/22606668/
Picot, A., Charbonnier, S., & Caplier, A. (2011). EOG-based drowsiness detection:
Comparison between a fuzzy system and two supervised learning classiers. IFAC
Proceedings Volumes, 44(1), 14283-14288. https://doi.org/10.3182/20110828-6-
IT-1002.00706
Picot, A., Charbonnier, S., & Caplier, A. (2012). On-Line Detection of Drowsiness
using Brain and Visual Information. IEEE Transactions on Systems, Man, and Cybernetics
- Part A: Systems and Humans, 42(3), 1083-4427. https://ieeexplore.ieee.org/
document/6029307
Zhang, C., Wang, H., & Fu, R. (2014). Automated Detection of Driver Fatigue Based on
Entropy and Complexity Measures. IEEE transactions on intelligent transportation systems,
15(1). https://ieeexplore.ieee.org/document/6589164
