EEG-based Person Authentication Using Multi-objective Flower Pollination Algorithm

Abstract

Since the past decades, the world has been transformed into a digital society, where every individual is living with a unique identifier. The primary purpose of this id is to distinguish from others and to deal with digital machines which are surrounding the world. Recently, many researchers showed that the brain electrical activity or electroencephalogram (EEG) signals could provide robust and unique features that can be considered as a new biometric authentication technique, given that accurately methods to decompose the signals must also be considered. This paper proposes a novel method for EEG signal denoising based on the multi-objective Flower Pollination Algorithm and the Wavelet Transform (MOFPA-WT) to extract useful features from denoised signals. MOFPA-WT is tested using a standard EEG signal dataset, namely, EEG motor movement/imagery dataset, and its performance is evaluated using three criteria: (i) accuracy, (ii) true acceptance rate, and (iii) false acceptance rate. We show that the proposed method can achieve results that are comparable to the state-of-the-art ones, as well as we draw future directions towards the research area.

Full text

EEG-based Person Authentication Using
Multi-objective Flower Pollination Algorithm
Zaid Abdi Alkareem Alyasseri∗† , Ahamad Tajudin Khader∗, Mohammed Azmi Al-Betar‡, Jo˜
ao P. Papa§,
Osama Ahmad Alomari∗
∗School of Computer Sciences, Universiti Sains Malaysia, Pulau Pinang, Malaysia
†ECE Department-Faculty of Engineering, University of Kufa, Najaf, Iraq
‡Department of Information Technology, Al-Huson University College, Al-Balqa Applied University, Al-Huson, Irbid, Jordan
§S˜
ao Paulo State University, Department of Computing, Bauru, Brazil
Abstract—Since the past decades, the world has been trans-
formed into a digital society, where every individual is living
with a unique identifier. The primary purpose of this id is to
distinguish from others and to deal with digital machines which
are surrounding the world. Recently, many researchers showed
that the brain electrical activity or electroencephalogram (EEG)
signals could provide robust and unique features that can be
considered as a new biometric authentication technique, given
that accurately methods to decompose the signals must also
be considered. This paper proposes a novel method for EEG
signal denoising based on the multi-objective Flower Pollination
Algorithm and the Wavelet Transform (MOFPA-WT) to extract
useful features from denoised signals. MOFPA-WT is tested
using a standard EEG signal dataset, namely, EEG motor
movement/imagery dataset, and its performance is evaluated
using three criteria: (i) accuracy, (ii) true acceptance rate, and
(iii) false acceptance rate. We show that the proposed method
can achieve results that are comparable to the state-of-the-art
ones, as well as we draw future directions towards the research
area.
Index Terms—EEG, Biometric, Authentication, Flower polli-
nation algorithm, multi-objective
I. INTRODUCTION
Electroencephalogram (EEG) is a graphical recording of
brain electrical activity that is recorded from the scalp, and it
stands for the voltage fluctuations resulting from ionic current
flows within the neurons of the brain [22], [33]. Therefore,
EEG signals can provide most of the required information
about brain activity, and they are acquired using invasive or
non-invasive techniques [26]. The main difference between
these methods is that the invasive approach involves the use of
electrode arrays implanted inside the brain, such as ECoG BCI
for arm movement control [27]. Berger [13] proposed the first
use of EEG signals as a non-invasive technique for capturing
brain activities. Over the past several decades, researchers have
developed Hans’s technique to suit versatile applications. For
example, EEG signals have been used in medical applications
for prevention, diagnosis, rehabilitation, and restoration of
patients. The EEG has also been used for non-medical appli-
cations, such as education and self-regulation, neuromarketing
Corresponding author: Zaid Abdi Alkareem Alyasseri (email:
zaid.alyasseri@uokufa.edu.iq
and advertisement, neuroergonomics and smart environment,
games and entertainment, and learning and education, as
summarized in [1]. Recently, EEG signals have been suc-
cessfully used as a new biometric technique in security and
authentication applications [1], [21], [22]. The researchers
found that EEG signals provide robust and unique features
that can be considered as a new biometric technique, being
also laborious to be spoofed [26].
Kumar et al. in [22] proposed a user identification system
based on EEG signal collected from six users using EMOTIVE
EPOC headset with 14 channels. In the preprocessing phase,
they used a Butterworth 5th-order filter with a range of 6-
35 Hz to achieve the highest signal-to-noise ratio (SNR)
of the input EEG signal. In the feature extraction phase,
Wavelet Transform (WT) is used to extract features of the
EEG signal. Also, three basic statistical measurements (i.e.,
mean, standard deviation, and energy) were extracted from
EEG signals. Concerning the classification phase, LVQ-NN
was used to recognize the users. Finally, the recognition rate
has been calculated over the different scenarios to find the
best combination of channels that can provide the accurate
classification.
Later, the same authors investigated some cognitive tasks
to design an individual identification system [33]. They used
standard EEG datasets related to motor/movement and imagi-
nary tasks [32] with one channel only (i. e., Cz) to obtain the
input signal. Also, the authors employed WT to decompose the
EEG signal into five levels to extract four different features
from each sub-band, namely: (i) energy, (ii) logarithm energy,
(iii) absolute energy, and (iv) REE energy. A neural network
classifier was proposed to classify EEG signals from five users
under four different train-test scenarios based on two tasks.
The authors found that the highest identification rates can be
obtained using the cognitive tasks based on motor imagination
when compared with the results based on motor movement.
Zahhad et al. [2] introduced a new method to improve the
performance of EEG-based biometric authentication systems
using eye-blinking electrooculographic (EOG) signals. Ro-
drigues et al. [28] used a binary-constrained Flower Pollination
Algorithm [35] to learn the best channels that can provide the

highest recognition rate for person identification based on EEG
signals. Their work was tested using standard EEG datasets
as well [32], which recognition rates nearly to 87%, while
reducing the number of EEG channels to half.
Kumar et al. [23] proposed a novel method for EEG feature
generation that applies a canonical correlation analysis for
feature fusion and further improving the recognition rate of
the identification technique. The proposed method was tested
using a standard EEG dataset [19] that has five mental tasks:
(i) baseline, (ii) multiplication of two numbers, (iii) geometric
figure rotation, (iv) letter composing, and (v) visual counting.
Each task was repeated several times for ten seconds and
the EEG signals collected from seven subjects. The proposed
method used three techniques to extract the features from the
input EEG signals: (i) empirical mode decomposition, (ii) in-
formation theoretic measure, and (iii) statistical measurement.
A linear vector quantization (LVQ) neural network and its
extension (LVQ2) have been used for classification purposes
on different mental tasks. The performance of the proposed
method provided better results compared with a na¨
ıve method
for feature fusion.
Safont et al. [29] proposed a biometric authentication
method using EEG signals that uses three EEG channels to
capture self-collected data from 70 subjects. The authors tested
the proposed system using six classification methods, achiev-
ing an equal error rate as of 2.4, as well as a classification
rate as of 93.8%. Jayarathne et al. [18] proposed a novel
approach for EEG-based on biometric authentication as well.
They suggested using the EEG signal rather than PIN number
for authenticating a person when using ATMs.
Although some works make use of meta-heuristic-based
optimization techniques in the context of EEG-based identi-
fication, we have not observed any work that makes use of
multi-objective optimization. Therefore, the main objective of
this paper is to propose a multi-objective Flower Pollination
Algorithm combined with Wavelet Transform (MOFPA-WT)
to decompose the input EEG signal and find the features that
can achieve the highest accuracies. MOFPA-WT is designed
using two objective functions: min(MSE) and max(SNR) to
obtain the best combination of WT parameters for EEG signal
denoising. The proposed method is implemented according to
the weighted-sum approach to combine multi-objectives into
a composite one objective function.
In this work, we used the standard Motor Move-
ment/Imagery dataset1EEG dataset [17], which includes 109
volunteers with EEG signals captured from 64 channels based
on different cognitive tasks. The original EEG signals are then
decomposed into five levels to extract features from each sub-
band (i.e., high gamma, gamma, alpha, beta, theta, and delta),
where four features are considered: energy, logarithm energy,
absolute energy, and REE energy [33].
To evaluate the performance of MOFPA-WT, the results are
assessed regarding three measurement factors: accuracy, true
acceptance rate, and false acceptance rate. The remainder of
1https://www.physionet.org/physiobank/database/eegmmidb/
this paper is organized as follows. Section II describes the
basis for WT-based EEG signal denoising. Section III provides
background about the Flower Pollination Algorithm and its
multi-objective variant. Section IV describes the proposed
system, and the results are discussed in Section V. Finally,
the conclusion and future works are stated in Section VI.
II. EEG SIGNAL DENOISING USING WAVEL ET
TRANSFORM
The Wavelet Transform is a powerful and widespread tool
for time-frequency domain signal representation, and it has
successfully applied for signal compression, feature extraction,
and selection, among others [6]–[9], [20]. In general, the WT
can be classified into two types: discrete wavelet transform
(DWT) and continuous wavelet transform (CWT) [31]. Re-
cently, the WT has been extensively used with non-stationary
signals, such as ECG and EEG, mainly due to its robust
outcomes in removing EEG artifact noises [20].
In this paper, the DWT has been used to decompose the
input EEG signal to extract unique features from each EEG
sub-band (i.e., high gamma, gamma, alpha, beta, theta, and
delta). One of the popular methods for DWT concerns so-
called “Donoho’s” approach, which extracted coefficients as
follows [34]:
C(a, b) =
N
X
i=1
s(i)gj,k(i)(1)
where C(a, b)denotes the wavelet dynamic coefficients, a=
2−jand b=k2−j, such that j∈ Z,k∈ Z, and Nstands
for the length of the signal. Additionally, astands for the size
of the time scale, bdenotes the translation, s(n)is the input
EEG signal, and gj,k(n)=2j/2g(2jn−k)is the DWT.
The task of DWT is to decompose the input signal using
different coefficient levels to correct the high frequency of the
input signal. The denoising process involves three phases:
•EEG signal decomposition: the original EEG signal is
divided into five levels, at each level the EEG signal
is decomposed into two parts, namely approximation
coefficients (cA), and detail coefficients (cD). The latter
step comprises using a high-pass filter over the signal,
and cA will be continuously decomposed for next level.
•Thresholding: for each level, a threshold value is defined
according to the noise level of the coefficients.
•Reconstruction: the EEG denoised signal is reconstructed
using inverse discrete wavelet transform iDWT.
The WT has five parameters with different types (Table I).
The efficiency of noise reduction and unique features ex-
traction relies on the selection of the wavelet parameters.
The wavelet denoising process has three phases: the first one
concerns the decomposition of the EEG signal using DWT.
This phase involves selecting the appropriate mother wavelet
function (Φ) to be used in the EEG signal decomposition
task. The second wavelet parameter, i.e., the decomposition
level (L), is also selected in this phase, and it requires some
expertize and experience. It should be noted that the selection

of appropriate parameters of the Wavelet Transform (which is
one of the main goals of this paper) is recently accomplished
using optimization techniques, such FPA, β-hill climbing (β-
hc), and genetic algorithm (GA) [7], [9].
In the second phase, the thresholding is applied. The
wavelet provides two standard types of thresholding functions
(β), namely hard and soft thresholding [14], [15]. The thresh-
olding type (soft or hard), selection rules (λ), and rescaling
methods (ρ) must all be selected as well, and they strongly
affect the global denoising performance. The thresholding
value is generally defined based on the standard deviation (σ)
of the amplitude the signal [16]. The wavelet parameters (i.e.,
β,λ, and ρ) must be separately applied for each level (i.e.,
cA and cD) . In the last phase, the denoised EEG signal is
reconstructed by iDWT.
TABLE I: Intervals of the WT denoising parameters.
Wavelet denoising parameters Method (range)
Wavelet function Φ
Symlet (sym1..sym45), Coiflet (coif1..coif5),
Daubechies (db1..db45), and Biorthogonal
(bior1.1.. bior1.5&bior2.2 .. bior2.8&bior3.1..bior3.9).
Thresholding function βsoft or hard
Decomposition level L5
Thresholding selection rule λHeursure, Rigsure, Sqtwolog, and Minimax
Rescaling approach ρone, sln, and mln
III. BACKGRO UN D
This section provides a background concerning the Flower
Pollination Algorithm and its multi-objective version.
A. Flower Pollination Algorithm
In a nutshell, meta-heuristic techniques can be classified
into evolutionary algorithms [5], [12], swarm intelligence
[10], and trajectory-based algorithms [3], [4], [8]. The Flower
Pollination Algorithm is inspired by the pollination behavior of
the flowering plants. FPA was introduced by Yang in 2012 [35]
and successfully applied for many optimization problems [10],
[25]. The rules (operators) of FPA are summarized as follows:
•Rule(1): global pollination involves both the biotic and
cross-pollination steps, where the pollinators carry the
pollen based on L´
evy flights.
•Rule(2): local pollination involves abiotic and self-
pollination.
•Rule(3): the reproduction probability is proportional to
the similarity between any two flowers.
•Rule(4): the switch probability p∈[0,1] can be con-
trolled between local and global pollination. Due to some
external factors (e.g., the wind), the local pollination
comprises a significant portion of the overall pollination
activities.
The key rules can be summarized in the pseudocode of the
FPA implemented in Algorithm 1, where ∈ < denotes a
small amount of value.
B. Multi-objective Optimization
This section describes a brief introduction to multi-objective
optimization. In general, such techniques refer to solving
Algorithm 1 Flower Pollination Algorithm pseudo-code
1: Objective: min f(x),x∈ <d
2: Initialize a population of nflowers (pollens) with random solutions.
3: Find the best solution x∗in the initial population.
4: Define a switch probability p∈[0,1].
5: Calculate all (f(x)) for nsolutions.
6: t= 0
7: while t≤MaxGeneration do
8: for i= 1, .., n do
9: rnd ← U (0,1).
10: if rnd ≤pthen
11: Draw a d-dimensional step vector ξwhich obeys a Le´
evy distribution.
12: Perform global pollination via xt+1
ixt
iξ∗(g∗−xt
i).
13: else
14: Draw from a uniform distribution U(0,1).
15: Randomly choose jand kamong all solutions, such that j6=k.
16: Perform local pollination via xt+1
i=xt
i+(xt
j−xt
k).
17: end if
18: Calculate f(x0)).
19: if f(x0)≤f(x)then
20: x←x0.
21: end if
22: end for
23: Find the current best solution x∗among all xt
i.
24: t=t+ 1.
25: end while
any optimization problem using more than one objective
function [24], [37]. The multi-objective optimization problem
for mobjective functions can be formulated as follows:
Min f(x) = {f1(x), f2(x), . . . , fm(x)}.(2)
where nrefers to number of objective functions.
The FPA technique has been extended to the multi-objective
optimization domain by Yang et al. [36], while the au-
thors adapted multi-objective Flower Pollination Algorithm
(MOFPA) for solving engineering optimization problems.
MOFPA is implemented according to the weighted-sum ap-
proach to combine two objectives into a composite one objec-
tive function. MOFPA defines the multi-objective optimization
as follows:
f=
M
X
k=1
Wkfk(3)
and m
X
k=1
Wk= 1, Wk>0,(4)
where Wkstand for a non-negative weight. The essential
idea of the weighted-sum approach is that these weighting
coefficients consider the preferences for the multi-objectives.
Given a set of weights, there is a single Pareto front that is
going to be generated for each weight.
IV. PROPOSED APPROACH
In this section, we introduce the proposed system for
EEG-based user authentication. The proposed approach goes
through four phases, in which the output of each stage works
as an input to the consecutive one. Figure 1 depicts the
proposed approach, which is detailed next.

Fig. 1: EEG-based user authentication system proposed in this wok.
A. EEG signal acquisition
The EEG signal acquisition is performed over a standard
EEG signal dataset [17]. The EEG signals are collected
from 109 healthy subjects using a brain-computer interface
software called BCI2000 system [32]. The EEG signals are
then recorded using 64 electrodes (i.e., channels), and each
user performs several motor/imagery tasks that are mainly
used in different fields, such as neurological rehabilitation and
brain-computer interface applications. In general, these tasks
consist of imagining or simulating a given action, such as
opening and closing the eyes. The EEG signals are recorded
from each volunteer by asking them to perform two tasks
according to the position of a target that appears on the screen
placed in front of them, as follows:
•Task 1: if the target appears on the right or left side of the
screen, then the volunteer must open and close his/her fist
corresponding to the position of the target on the screen.
Then the volunteer relaxes.
•Task 2: if the target appears on the right or left side of
the screen, then the volunteer imagines open and close
his/her fist corresponding to the position of the target on
the screen. Then the volunteer relaxes.
In this paper, the input EEG signals are collected only from the
cerebral signal (Cz channel) because the cerebral region has
a high activity when the user performs any motor-movement
task [33]. The acquisition of the input EEG signal is repeated
for each user three times, with one minute for interval. Later
on, the signal is divided into six samples with 10seach.
Fig. 2: EEG-based user authentication system proposed in this
wok.
B. EEG signal denoising using MOFPA-WT
Despite WT has many advantages and has been successfully
used for denoising non-stationary signals such as ECG and
EEG [7], [9], most of the current approaches degrade the
energy of the original signal when reducing its noise. This
situation usually occurs because they consider only the mini-
mum squared error (MSE) between the original and denoised
signals. For that reason, this work designs a multi-objective

function that considers a balance between reducing the EEG
noise and keeping its signal energy.
In this paper, we propose to estimate the optimum/near-
optimum set of parameters concerning the Wavelet Transform
for EEG signal denoising as a multi-objective optimization
task. In our approach, the set of WT parameters is represented
as a vector x= (x1, x2, . . . , xd)where dis the number of pa-
rameters used for the Wavelet Transform2. In this context, x1
represents the value of the mother wavelet function parameter
Φ,x2stands for the value of the decomposition level parameter
L,x3refers to the thresholding method β,x4represents the
value of the thresholding selection rule parameter λ, and x5
represents the re-scaling approach ρ(the possible ranges for
these parameters are described in Table I). Figure 3 depicts
the representation of a possible solution using the proposed
approach.
Fig. 3: Modeling the problem of WT configuration for EEG
signal denoising using MOFPA-WT.
The proposed MOFPA-WT evaluates each solution using the
multi-objective framework applying two objective functions:
min(MSE) and max(SNR), as formulated below:
f=W1f1+W2f2(5)
=W1∗min(MSE ) + W2∗max(SNR),
(6)
where the weight vector is initialized as follows:
W1∼U(0,1),(7)
W2= 1 −W1.
The two objective functions which are mean squared error
(MSE) and signal-to-noise ratio (SNR) are formulated as
below:
MSE =1
N
N
X
i=1
[x(i)−bx(i)]2(8)
and
SN R = 10 log10 (PN
i=1[x(i)]2
PN
i=1[x(i)−bx(i)]2),(9)
2In this paper, d= 5.
where x(i)and bx(i)denote the original and denoised EEG
signals, respectively. Notice that bx(i)is obtained using the
Wavelet Transform tuned by the proposed MOFPA-WT.
Iteratively, the randomly generated solutions undergo refine-
ment using the MOFPA-WT. The final result of this phase is an
optimized solution x∗= (x∗
1, x∗
2, . . . , x∗
d)that will be passed
to the denoising phase, which involves three main steps that
are depicted in Figure 4 and described in more details below:
•EEG signal decomposition using DWT: in this step, the
DWT is applied to decompose the noise of the input
EEG signals. In such process, one uses the first two
x∗parameters (i.e., the mother wavelet function and the
decomposition level) only. Figure 4 shows the DWT pro-
cedure for five levels, where the EEG signal is partitioned
at each level into cA and cD components. The latter is
processed using a high-pass filter, while the former is
processed using a low-pass filter and is decomposed for
the next level.
•Thresholding: it is applied based on the noise level of the
coefficients. In this step, the last three wavelet parame-
ters, namely, the thresholding type (β), the thresholding
selection rules (λ), and the re-scaling methods (ρ), must
be selected from x∗.
•Reconstruction of the denoised EEG signal by iDWT:
we estimate the value of the original EEG signals by
applying iDWT on their denoised version. The recon-
struction convolves the EEG data using upsampling,
which involves the addition of zeros at the even index
elements of the signal.
C. Feature Extraction
Extracting effective features plays a significant role in any
authentication system [11], [30], [33]. Therefore, the main
purpose of this phase is to find unique information from
each sub-band (i.e., high gamma, gamma, alpha, beta, theta,
and delta) that allow the classification to obtain good results.
Figure 4 shows the feature extraction adopted in this work
that is based on WT decomposition with five levels. Table II
presents the characteristics of the EEG rhythms.
There are several features that can be extracted from the
denoised EEG signal. In this paper, we applied four variants
of the energy that can be obtained through the sub-bands:
feature1=EEGE nergy =
M
X
j=1
|wij |2, i = 1,2,3, ..., L
(10)
where Mstands for the number of coefficients wij. The next
features are calculated as follows:
feature2=Energy of EEG Rhythm
EEGEnergy
∗100,(11)
feature3= log(f eature2),(12)
and

feature4=Abs(f eature3).(13)
Fig. 4: EEG feature extraction based WT decomposition with
five levels.
TABLE II: EEG rhythm main information.
Rhythm Frequency Range Frequency bandwidth (Hz) Decomposition level
Delta(δ) 0-4(Hz) 4 Level 5 (Ac 5)
Theta(θ) 4-8(Hz) 4 Level 5 (Dc 5)
Alpha(α) 8-13(Hz) 5 Level 4 (Dc 4)
Beta(β) 14-30(Hz) 16 Level 3 (Dc 3)
Low Gamma(γ) 30-64(Hz) 34 Level 2 (Dc 2)
High Gamma(γ) 64-128(Hz) 64 Level 1 (Dc 1)
Table III presents the number of features that are used in
this work.
D. Neural Network classifier
An artificial neural network (ANN) classifier was applied to
classify the extracted features from the denoised EEG signal
into correct persons. We designed a network with 24-input
features from each subject (i.e., 4features ∗6sub-bands), 32
hidden layers, and 5output layers (i.e., the proposed system on
evaluated on five users). Notice that such neural architecture
was empirically chosen.
V. RE SU LTS AND DISCUSSIONS
The EEG dataset used in this paper was divided into four
different scenarios based on the standard training-and-testing
approach for each task. To evaluate the performance of the
proposed MOFPA-WT, we considered three measures, namely
accuracy, true acceptance rate (TAR), and false acceptance rate
(FAR) which can formulated as follows:
Accuracy =T A +T R
T A +F A +T R +F R ,(14)
Sensitivity(T AR) = T A
T A +F R ,(15)
Fig. 5: Multi-layer back propagation ANN adopted in this
work.
Specif ity(TFR) = T R
T R +F R ,(16)
and
F AR = 1 −T F R, (17)
where TA, TR, FA, and FR represent the true acceptance,
true reject, false acceptance, and false reject, respectively.
The results of the classification phase are represented as a
confusion matrix that tabulates whether they fall into one of
four categories: TA, TR, FA and FR.
Table IV presents the experiment concerning the training
step with 11 persons and testing with 7individuals (task 1).
In this case, we obtained a TAR as of 85.71% and FAR as
of 14.28%, which are pretty much interesting and suitable
for EEG-based people identification. Table V presents the
experiment concerning the training step with 10 persons and
testing with 8individuals (task 1). In this case, we obtained a
TAR as of 90% and FAR as of 10%, i.e., we obtained better
recognition rates when compared to the previous experiment
but using less training cases.
With respect to the task 2, Table VI presents the experiments
using 11 individuals for training and 7for testing purposes.
In this case, we obtained a TAR value as of 91.42% and FAR
as of 8.58%. Table VII presents the experiments using 10
individuals for training and 8for testing purposes. In this case,
we obtained a TAR value as of 85% and FAR as of 15%.
In a nutshell, the best recognition rates were obtained in
the task 2 scenario and using 11 individuals for training
and 7for testing purposes. Task 1 obtained better results
with less training samples, but the opposite can be observed
concerning task 2. However, it is not possible to state whether
such assumptions are significant due to the small number
of individuals in the dataset. Even nowadays, it is still not
straightforward to obtain large datasets for EEG-based person
identification. We believe that in the nearby future, small and
portable devices for EEE signal acquisition may be available
at affordable prices.
Figure 6 shows the accuracy rate considering the input EEG
signals based on five decomposition levels using MOFPA-WT.
For the sake of visualization purposes, this chart summarizes

TABLE III: No. of features of the EEG dataset.
No. of subjects (X) No. of samples (S) No. of trails (Tr) No. of tasks (Ta) No. sub-band no. of features Total No. of features
5 6 3 2 6 4 4320 features
TABLE IV: Confusion matrix concerning the experiment with
11 persons for training and 7for testing (task 1).
Sub1 Sub2 Sub3 Sub4 Sub5 Specificity
Sub1 70000 0.928
Sub2 0 7000 1
Sub3 2 0 50 0 0.933
Sub4 0 0 0 520.9
Sub5 0 0 0 1 6 0.896
FAR 0 0 2/7 2/7 1/7
TABLE V: Confusion matrix concerning the experiment with
10 persons for training and 8for testing (task 1).
Sub1 Sub2 Sub3 Sub4 Sub5 Specificity
Sub1 70100 0.968
Sub2 0 8000 1
Sub3 1 0 70 0 0.939
Sub4 0 0 0 710.939
Sub5 0 0 0 1 7 0.939
FAR 1/8 0 1/8 1/8 1/8
the experiments conducted in this work. One can observe that
the multi-objective paradigm is quite promising to be applied
in the context of EEG signal denoising based on the Wavelet
Transform. The results obtained are pretty much interesting
and they can be extended to larger datasets when they happen
to be available. With such a combined framework, we can
learn, simultaneously, how to reconstruct and denoise the
signal.
VI. CONCLUSIONS AND FUTURE WORK
In this paper, a novel technique for EEG signal denoising
based on multi-objective Flower Pollination Algorithm with
wavelet transform was proposed. The main task of MOFPA-
WT method is to find the efficient decomposition of the input
TABLE VI: Confusion matrix concerning the experiment with
11 persons for training and 7for testing (task 2).
Sub1 Sub2 Sub3 Sub4 Sub5 Specificity
Sub1 70000 0.964
Sub2 0 7000 0.964
Sub3 1 0 60 0 0.965
Sub4 0 0 0 610.965
Sub5 0 1 0 0 6 0.931
FAR 0 0 1/7 1/7 1/7
TABLE VII: Confusion matrix concerning the experiment with
10 persons for training and 8for testing (task 2).
Sub1 Sub2 Sub3 Sub4 Sub5 Specificity
Sub1 60200 0.968
Sub2 0 8000 1
Sub3 1 0 70 0 0.909
Sub4 0 0 0 530.914
Sub5 0 0 0 0 8 0.906
FAR 2/8 0 1/8 3/8 0
EEG signal which can provide unique features from each sub-
bands. MOFPA-WT is tested using a standard EEG signal
dataset, namely, EEG motor movement/imagery dataset. The
performance of MOFPA-WT is evaluated using three criteria,
namely, accuracy, TAR, and FAR. The proposed method
achieved the highest accuracies using the cognitive tasks based
on motor movement compared with the results based on motor
imagination.
Regarding future works, we intend to apply MOFPA-WT
in more challenging signal problem instances, such as user
authentication or early detection of epilepsy based on EEG
signals, as well as to consider MOFPA-WT in larger datasets
for EEG-based person identification.
ACKNOWLEDGMENT
The first author would like to thank the University Sci-
ence Malaysia (USM) and The World Academic Science
(TWAS) for supporting his PhD study which is a USM-
TWAS Postgraduate Fellowship, FR number: 3240287134.
The fourth author is grateful to FAPESP grants #2013/07375-
0, #2014/12236-1, and #2016/19403-6, as well as CNPq grants
#306166/2014-3 and #307066/2017-7. This material is based
upon work supported in part by funds provided by Intel AI
Academy program under Fundunesp Grant No.2597.2017.
REFERENCES
[1] S. N. Abdulkader, A. Atia, and M.-S. M. Mostafa. Brain computer
interfacing: Applications and challenges. Egyptian Informatics Journal,
16(2):213–230, 2015.
[2] M. Abo-Zahhad, S. M. Ahmed, and S. N. Abbas. A new multi-level
approach to eeg based human authentication using eye blinking. Pattern
Recognition Letters, 82:216–225, 2016.
[3] L. M. Abualigah, A. T. Khader, M. A. Al-Betar, Z. A. A. Alyasseri,
O. A. Alomari, and E. S. Hanandeh. Feature selection with β-hill
climbing search for text clustering application. In Information and
Communication Technology (PICICT), 2017 Palestinian International
Conference on, pages 22–27. IEEE, 2017.
[4] M. A. Al-Betar. β-hill climbing: an exploratory local search. Neural
Computing and Applications, pages 1–16, 2016.
[5] M. A. Al-Betar, Z. A. A. Alyasseri, A. T. Khader, A. L. Bolaji, and
M. A. Awadallah. Gray image enhancement using harmony search.
International Journal of Computational Intelligence Systems, 9(5):932–
944, 2016.
[6] Z. A. A. Alyasseri, A. T. Khader, and M. A. Al-Betar. Electroen-
cephalogram signals denoising using various mother wavelet functions:
A comparative analysis. In ICISPC ’17, Penang, Malaysia, pages 66–70.
ACM, 2017.
[7] Z. A. A. Alyasseri, A. T. Khader, and M. A. Al-Betar. Optimal eeg
signals denoising using hybrid β-hill climbing algorithm and wavelet
transform. In ICISPC ’17, Penang, Malaysia, pages 147–152. ACM,
2017.
[8] Z. A. A. Alyasseri, A. T. Khader, M. A. Al-Betar, and L. M. Abualigah.
Ecg signal denoising using β-hill climbing algorithm and wavelet trans-
form. In ICIT 2017 The 8th International Conference on Information
Technology, pages 1–7, 2017.
[9] Z. A. A. Alyasseri, A. T. Khader, M. A. Al-Betar, and M. A. Awadallah.
Hybridizing β-hill climbing with wavelet transform for denoising ecg
signals. Information Sciences, 429:229–246, 2018.

Fig. 6: Accuracy results based on five decomposition levels using MOFPA-WT.
[10] Z. A. A. Alyasseri, A. T. Khader, M. A. Al-Betar, M. A. Awadallah,
and X.-S. Yang. Variants of the flower pollination algorithm: A review.
In Nature-Inspired Algorithms and Applied Optimization, pages 91–118.
Springer, 2018.
[11] Z. A. A. Alyasseri, A. T. Khader, M. A. Al-Betar, J. P. Papa, A. A.
Osama, and S. N. Makhadme. An efficient optimization technique of
eeg decomposition for user authentication system. In 2nd International
Conference on BioSignal Analysis, Processing and Systems (ICBAPS),
pages 25–31. IEEE, 2018.
[12] Z. A. A. Alyasseri, I. Venkat, M. A. Al-Betar, and A. T. Khader. Edge
preserving image enhancement via harmony search algorithm. In Data
Mining and Optimization (DMO), 2012 4th Conference on, pages 47–52.
IEEE, 2012.
[13] H. Berger. ¨
Uber das elektrenkephalogramm des menschen. European
Archives of Psychiatry and Clinical Neuroscience, 87(1):527–570, 1929.
[14] D. L. Donoho. De-noising by soft-thresholding. IEEE transactions on
information theory, 41(3):613–627, 1995.
[15] D. L. Donoho and J. M. Johnstone. Ideal spatial adaptation by wavelet
shrinkage. Biometrika, 81(3):425–455, 1994.
[16] E.-S. A. El-Dahshan. Genetic algorithm and wavelet hybrid scheme for
ecg signal denoising. Telecommunication Systems, 46(3):209–215, 2011.
[17] A. L. Goldberger, L. A. Amaral, L. Glass, J. M. Hausdorff, P. C. Ivanov,
R. G. Mark, J. E. Mietus, G. B. Moody, C.-K. Peng, and H. E. Stanley.
Physiobank, physiotoolkit, and physionet. Circulation, 101(23):e215–
e220, 2000.
[18] I. Jayarathne, M. Cohen, and S. Amarakeerthi. Brainid: Development
of an eeg-based biometric authentication system. In Information Tech-
nology, Electronics and Mobile Communication Conference (IEMCON),
2016 IEEE 7th Annual, pages 1–6. IEEE, 2016.
[19] Z. A. Keirn and J. I. Aunon. A new mode of communication between
man and his surroundings. IEEE transactions on biomedical engineer-
ing, 37(12):1209–1214, 1990.
[20] H. Kumar, S. P. Pai, G. Vijay, and R. Rao. Wavelet transform for bearing
condition monitoring and fault diagnosis: A review. International
Journal of COMADEM, 17(1):9–23, 2014.
[21] P. Kumari and A. Vaish. Brainwave based authentication system:
research issues and challenges. International Journal of Computer
Engineering and Applications, 4(1):2, 2014.
[22] P. Kumari and A. Vaish. Brainwave based user identification system: A
pilot study in robotics environment. Robotics and Autonomous Systems,
65:15–23, 2015.
[23] P. Kumari and A. Vaish. Feature-level fusion of mental tasks brain
signal for an efficient identification system. Neural Computing and
Applications, 27(3):659–669, 2016.
[24] S. Mirjalili, S. Saremi, S. M. Mirjalili, and L. d. S. Coelho. Multi-
objective grey wolf optimizer: a novel algorithm for multi-criterion
optimization. Expert Systems with Applications, 47:106–119, 2016.
[25] A. A. Osama, A. T. Khader, M. A. Al-Betar, and Z. A. A. Alyasseri.
A hybrid filter-wrapper gene selection method for cancer classification.
In 2nd International Conference on BioSignal Analysis, Processing and
Systems (ICBAPS), pages 32–37. IEEE, 2018.
[26] R. A. Ramadan and A. V. Vasilakos. Brain computer interface: control
signals review. Neurocomputing, 223:26–44, 2017.
[27] R. P. Rao. Brain-computer interfacing: an introduction. Cambridge
University Press, 2013.
[28] D. Rodrigues, G. F. Silva, J. P. Papa, A. N. Marana, and X.-S. Yang. Eeg-
based person identification through binary flower pollination algorithm.
Expert Systems with Applications, 62:81–90, 2016.
[29] G. Safont, A. Salazar, A. Soriano, and L. Vergara. Combination of
multiple detectors for eeg based biometric identification/authentication.
In Security Technology (ICCST), 2012 IEEE International Carnahan
Conference on, pages 230–236. IEEE, 2012.
[30] N. D. Sarier. Improving the accuracy and storage cost in biometric
remote authentication schemes. Journal of Network and Computer
Applications, 33(3):268–274, 2010.
[31] C. Sawant and H. T. Patii. Wavelet based ecg signal de-noising.
In Networks & Soft Computing (ICNSC), 2014 First International
Conference on, pages 20–24. IEEE, 2014.
[32] G. Schalk, D. J. McFarland, T. Hinterberger, N. Birbaumer, and J. R.
Wolpaw. Bci2000: a general-purpose brain-computer interface (bci)
system. IEEE Transactions on biomedical engineering, 51(6):1034–
1043, 2004.
[33] P. K. Sharma and A. Vaish. Individual identification based on neuro-
signal using motor movement and imaginary cognitive process. Optik-
International Journal for Light and Electron Optics, 127(4):2143–2148,
2016.
[34] B. N. Singh and A. K. Tiwari. Optimal selection of wavelet basis
function applied to ecg signal denoising. Digital Signal Processing,
16(3):275–287, 2006.
[35] X.-S. Yang. Flower pollination algorithm for global optimization. In
International Conference on Unconventional Computing and Natural
Computation, pages 240–249. Springer, 2012.
[36] X.-S. Yang, M. Karamanoglu, and X. He. Multi-objective flower
algorithm for optimization. Procedia Computer Science, 18:861–868,
2013.
[37] X.-S. Yang, M. Karamanoglu, and X. He. Flower pollination algorithm:
a novel approach for multiobjective optimization. Engineering Opti-
mization, 46(9):1222–1237, 2014.