Content uploaded by Debangana Das
Author content
All content in this area was uploaded by Debangana Das on May 02, 2024
Content may be subject to copyright.
19842 IEEE SENSORS JOURNAL, VOL. 23, NO. 17, 1 SEPTEMBER 2023
Optimization Techniques for a Voltammetric
Signal to Predict Green Tea Quality
Parameters Using MIP Electrode
Srikanta Acharya , Debangana Das , Shreya Nag , Soumen Mukherjee , Ajanto Kumar Hazarika ,
Santanu Sabhapondit, Bipan Tudu , Rajib Bandyopadhyay , and Runu Banerjee Roy
Abstract—In this treatise, an integration of feature trans-
formation, optimization, and prediction algorithm has been
proposed for voltammetric signal to improve the prediction
accuracy of an electrochemical system. A three electrode
voltammetric system comprises a Ag/AgCl reference elec-
trode, a platinum counter electrode, and a synthesized work-
ing electrode (WE). Two molecularly imprinted polymer (MIP)
electrodes have been used for gallic acid (GAL) and epicat-
echin (EC) detection in green tea samples using differential
pulse voltammetry (DPV). Discrete cosine transform (DCT)
data have been optimized by genetic algorithm (GA), bat
algorithm (BA), and whale optimization algorithm (WOA).
A significant improvement was obtained in the model pre-
diction accuracy using 83 and 129 features by GA optimized
dataset. Reduced datasets were then used for prediction models using the partial least square regression (PLSR) and
principal component regression (PCR). The root means square error of calibration (RMSEC) obtained from PLSR and
PCR is 0.253 and 0.094 and 0.239 and 0.088 for GAL and EC, respectively. Prediction accuracies obtained for GAL and
EC through PLSR and PCR are 92.24% and 96.24% and 97.95% and 97%, respectively.
Index Terms—Discrete cosine transform (DCT), epicatechin (EC), gallic acid (GAL), molecularly imprinted polymer
(MIP), optimization.
I. INTRODUCTION
THERE are numerous health-promoting chemicals found
in green tea, and a few of these components are gallic
acid (GAL), catechin (CAT), epigallocatechin gallate, caffeine,
Manuscript received 26 June 2023; accepted 16 July 2023. Date of
publication 27 July 2023; date of current version 31 August 2023. The
associate editor coordinating the review of this article and approving
it for publication was Prof. Rosario Morello. (Corresponding author:
Runu Banerjee Roy.)
Srikanta Acharya, Bipan Tudu, Rajib Bandyopadhyay, and
Runu Banerjee Roy are with the Department of IEE, Jadavpur Univer-
sity, Kolkata 700106, India (e-mail: acharya.srikanta73@gmail.com;
bipantudu@gmail.com; bandyopadhyay.rajib@gmail.com; runuroy@
yahoo.com).
Debangana Das is with the Department of ECE, Silicon Institute
of Technology, Bhubaneswar 751024, India (e-mail: debanganadas4@
gmail.com).
Shreya Nag is with the Department of Electronics and Communica-
tion Engineering, University of Engineering and Management, Kolkata
700160, India (e-mail: snshreya20@gmail.com).
Soumen Mukherjee is with the Department of Information Technology,
RCC Institute of Information Technology, Kolkata 700015, India (e-mail:
soumou601@gmail.com).
Ajanto Kumar Hazarika and Santanu Sabhapondit are with
the Toklaiu Tea Research Institute, Jorhat 785008, India (e-mail:
ajantohazarika65@gmail.com; santanusabhapondit1@gmail.com).
Digital Object Identifier 10.1109/JSEN.2023.3297140
and epicatechin (EC). GAL is the most abundant free phenolic
component in tea and the main contributor to several health
benefit activities, such as antibiotic, anti-inflammatory, and
anticancer [1],[2],[3],[4] consequences. Some of the conven-
tional methods for detecting GAL content in food and drinks
include reversed-phase high-performance liquid chromatogra-
phy [5], flow injection analysis [6], chemiluminescence [7],
thin-layer chromatography [8], and chronoamperometry [9].
Electrochemical detection of GAL has been proposed to
circumvent these limitations because of their inherent selectiv-
ity [10],[11],[12]. The molecularly imprinted polymer (MIP)
approach has been used to adsorb a particular molecule amid
a pool of other molecules [13]. Recently, interest in MIPs
as a specialized recognition material has increased. Target
analytes are covalently or noncovalently bound to a network of
monomers and crosslinkers during molecular imprinting. After
polymerization, the template is eliminated, leaving behind a
molecularly stacked polymer matrix. In general, molecular
recognition is essential to every compound’s adsorption pro-
cess [14],[15]. Several MIP electrodes have been developed
for the detection of tea chemicals, such as GAL, caffeine, EC,
theaflavin, CATs, and so on [16],[17],[18]. Metaheuristic
methods [19],[20],[21],[22],[23],[24],[25] are particularly
1558-1748 © 2023 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See https://www.ieee.org/publications/rights/index.html for more information.
Authorized licensed use limited to: Silicon Institute of Technology. Downloaded on January 25,2024 at 10:19:04 UTC from IEEE Xplore. Restrictions apply.
ACHARYA et al.: OPTIMIZATION TECHNIQUES FOR A VOLTAMMETRIC SIGNAL 19843
very useful for solving problems relating to optimization.
Thus, a metaheuristic-based model has been proposed in
this work, where three meta-heuristic data optimization tech-
niques, viz., genetic algorithm (GA), bat algorithm (BA), and
whale optimization algorithm (WOA), have been employed to
address the problem.
Kalathingal et al. [19], the process parameters of a fluidized
bed drier, which is used for the final drying of the tea
leaves, have been optimized by the integration of an arti-
ficial neural network (ANN) with GA. The best attributes
of several tea samples were used to construct the GA-SVM
model in real-time quick measuring method [20]. Using a
GA-optimized dataset, the model’s overall classification accu-
racy has increased. The grades of Keemun black tea have
been identified in another study using a cognitive spec-
troscopy approach [21]. Four different screening techniques,
including the successive projections algorithm (SPA), com-
petitive adaptive reweighted sampling (CARS), and shuffled
frog leaping algorithm (SFLA), have been proposed to find
the key features information of various tea samples. These
screening techniques aim to improve the system’s predic-
tion accuracy. BA is based on the echolocation features
of microbats and possesses many advantages as follows:
it provides very quick classification, quick convergence by
switching exploration to exploitation, automatic zooming, and
parameter control capability to different application areas [22].
In another work, the BA, WOA, and intense weed optimization
(IWO) algorithm—all of which were inspired by nature—
were used for optimization due to their benefits and distinctive
features [23]. WOA is a simple, fast, and high-accuracy
algorithm that is composed of minimal coding. Liu et al. [24]
build a photovoltaic power-based symmetric model to perform
accurately in different weather conditions, and the selection
of optimal features was proposed by two binary WOA algo-
rithms [25]. When input data are highly correlated, discrete
cosine transform (DCT) shows strong energy compaction
properties than discrete Fourier transformation (DFT). In the
study [26], five polymer-graphite composite electrodes of the
electronic tongue have been optimized by an ANN classifier
followed by the DCT technique. In this work, the MIP-GAL
and Q-IPG electrodes have been fabricated following similar
protocols as stated in [17] and [18]. After the fabrication of the
MIP-GAL electrode, it was imbibed in green tea samples and
subjected to differential pulse voltammetry (DPV) analysis.
The responses were recorded and transformed using the DCT
feature transformation technique. The low-dimensional feature
dataset was obtained by applying different data optimization
techniques. These optimized feature sets were then used for
developing prediction models using the partial least square
regression (PLSR) technique and principal component regres-
sion (PCR). The calibration models were developed using
the HPLC analysis results. The same techniques have been
employed on Q-IPG electrodes for the detection of EC in
green tea. The root mean square error of calibration (RMSEC)
obtained using PLSR was 0.253 for GAL and 0.094 for
EC detection. The RMSEC values obtained from PCR were
0.239 and 0.088 for GAL and EC detection, respectively.
The novelty of this work is to develop a metaheuristic-based
model that increases the overall prediction accuracy of an
electrochemical system by deploying optimized and reduced
datasets. The main focus of this work was to improve the
prediction accuracy using the optimized reduced datasets to
that obtained by using raw datasets in [17]. The results indicate
that the generalized prediction model developed in this work
revealed higher prediction accuracy (96.24%) for GAL detec-
tion than 88.97%, which has been obtained in [17]. To validate
the model, the dataset obtained in EC detection [18] was
used. The same feature transformation and data optimization
techniques have been deployed. The resultant dataset was then
subjected to the developed prediction model and a prediction
accuracy of 97.95% was obtained achieved instead of 94.54%
in the previous work.
II. MATERIALS AND METHODS
A. Chemical Reagents and Material
GAL, ethylene glycol dimethyl acrylate (EGDMA), graphite
powder (99%), itaconic acid (IA), acrylamide (AAM), maleic
acid (MA), CAT, and ascorbic acid (AA) were purchased
from Sigma Aldrich, India. Copper chloride (CuCl2), ethanol,
cetylpyridinium chloride (C21H38NCl), sodium hydroxide
(NaOH), and paraffin oil were purchased from Merck and
Company, India. Benzoyl peroxide (BP) was purchased from
Sisco Research Laboratories Pvt. Ltd., India. Deionized water
(resistance 18 M), which is used for cleaning the electrodes,
was taken from the Millipore water purification system.
B. Synthesis of CuO NPs, GAL Imprinted Polymer (MIP),
and Nonimprinted Polymer (NIP)
The synthesis of CuO NPs was carried out in the laboratory
by the sol–gel method. The stepwise synthesis of the CuO NPs,
MIP, and NIP is described in the work [17],[18].
C. Synthesis of Quercetin Imprinted Electrode (Q-IPG)
and NIP
The synthesis of Q-IPG material was carried out in the
laboratory. The detailed study and fabrication of the electrode
synthesis are elucidated in [18].
III. DATA ANALYSIS
The DPV responses were obtained using the MIP-GAL
electrode by varying the voltage from 0 to 1 V. The 179 data
points obtained in the study correspond to the values of current
in microamperes (µA) upon the variation of the voltage in
the said range. For each tea sample, eight repetitive DPV
responses were recorded. Thus, for a single variant of tea,
the dataset size was obtained as (179 ×8). Thus, the size
of the raw data matrix for ten samples on imbibing the
MIP-CuO electrode to the GAL samples was obtained as
[179 ×8×10], and these data were transformed using the
DCT technique. A set of feature optimization technique, viz.,
GA, BA, and WOA, have been explored here. The predictive
ability of the synthesized electrodes is evaluated by the PLSR
and PCR model.
A. Feature Transformation
1) Discrete Cosine Transformation:DCT is an ortho normal
time-series Fourier transform like DFT. It achieves better
energy concentration than DFT when the successive values are
correlated. During the transformation, it uses real numbers and
does not change the original vector length. The computational
complexity of this method is very low [27].
Authorized licensed use limited to: Silicon Institute of Technology. Downloaded on January 25,2024 at 10:19:04 UTC from IEEE Xplore. Restrictions apply.
19844 IEEE SENSORS JOURNAL, VOL. 23, NO. 17, 1 SEPTEMBER 2023
B. Feature Optimization
Feature optimization is a process that reduces the dimension
of input variables during the development of the predic-
tive model. It reduces computational modeling costs and
improves the performance of the model. Here, three different
nature-inspired feature optimization techniques, viz., GA, BA,
and WOA, have been used.
1) Genetic Algorithm:GA is a well-known metaheuristic
algorithm, where variable selection is based on the princi-
ples of genetics and natural selection. The primary elements
of this stochastic algorithm are chromosome representation,
fitness selection, and biological-inspired operators. Here, bio-
inspired operators are selection, mutation, and crossover. The
chromosomes of GA are the binary string format and fitness is
the assigned value for all the chromosomes in the population.
During this process, different numbers of runs are necessary,
and the best running result is selected to produce the final
model with the lowest RMSECV value [28],[29].
2) Binary BA:The BA is a nature-inspired algorithm based
on the echolocation behavior of microbats. It uses frequency
tuning for optimization and computational intelligence. The
position, velocity, and frequency of an artificial bat are mod-
ified by every iteration [30],[31]. In searching time, bats
move around the free space utilizing their position and velocity
vectors within the searching area. To solve the feature selection
and classification problem, the binary BA is developed, which
is a discrete version of the BA. In binary BA, the presence and
absence of features are denoted by ones and zeros. Bat moves
to a new position in the free space by inverting the different
numbers of bits; hence, it uses a different strategy to modify
their velocity and position.
3) Whale Optimization Algorithm:The WOA is also a nature-
inspired meta-heuristic algorithm. It imitates the behavior
of humpback whales and stimulates the bubble-net hunting
strategy. The bubble-net feeding is a unique behavior that
can only be provided by humpback whales. Here, spiral
bubble-net feeding maneuver is used to perform optimization.
The mathematical model of this algorithm can be developed by
encircling prey, spiral bubble-net feeding maneuver, and prey
searching. The advantages of this method are simplicity and
low cost. WOA performance is verified for real challenging
optimization problems [32].
C. Regression Models
Herein, a function is developed to describe the relationship
between independent and dependent variables. Two general-
ized prediction models, namely, PLSR and PCR, have been
developed to address the problem.
1) Partial Least Square Regression:PLSR employs a lin-
ear multivariate model to compare the performance of the
response matrix and the predictor matrix [33]. Leave one
out cross-validation (LOOCV) method has been assigned as
a performance metric for this study. It removes a set of
components whose predictors and response variables have a
maximum correlation. Different parametric values, viz., root
mean square error of validation (RMSEV), root mean square
error of prediction (RMSEP), and RMSEC, are used to opti-
mize the latent variables (LVs), and hence, the performance
of the model has been evaluated.
2) Principal Component Regression:Here, the PCs are
found and evaluated by deploying PCA on the data matrix.
Fig. 1. Response in green tea samples of (a) MIP-GAL electrode and
(b) Q-IPG electrode.
TABLE I
NUMBER OF FEATURES AFT ER IMPLEMENTATION
OF OPTIMIZATION TECHNIQUES
The LVs are being optimized by minimizing the linear model
parameters, viz., RMSEC, RMSEV, and RMSEP, which are
the best for good prediction [34].
IV. RESULTS AND DISCUSSION
The procedure of meta-heuristic model development for
optimizing the performance of the synthesized MIP electrode
in green tea classification has been discussed in this section.
Two synthesized MIP electrodes: MIP-GAL for GAL detection
and Q-IPG electrode for EC detection in green tea are con-
sidered for the case study of this model. Here, two phenolic
compounds of green tea, viz., GAL and EC, are detected by
the voltammetric method. The DPV plots of two synthesized
electrodes are obtained by applying the potential window
of (0–1 V) with scan rate 0.01–0.4 Vs−1and (0–1.2 V)
with 0.025–0.3 Vs−1, respectively. The DPV plots have been
depicted in Fig. 1(a) and (b). The resultant matrix size for two
synthesized electrodes and ten variants of green tea samples
with eight replicates are [179 ×8×10] and [239 ×8×10],
respectively, and are taken into account for further analysis
using chemometrics [35],[36].
A. Feature Data Transformation and Data Optimization
The dataset of size [179 ×8×10] has been transformed
by a powerful feature transformation technique DCT, which
employs the cosine function instead of sine and cosine both
functions. The coefficients of DCT are real and provide better
energy compaction capability when the dataset is highly cor-
related. Here, data optimization and dimension reduction have
been performed using very popular bio-inspired meta-heuristic
data optimization algorithms, viz., GA, BA, and WOA on the
DCT transformed feature.
1) Data Optimization Using GA:GA is employed on the
DCT transformed dataset of size [179 ×8×10]. Here, we use
0.1 as a mutation rate (µ), 0.7 as a crossover percentage
(pc), and 0.3 as a mutation percentage (pm) for the data
optimization. The cost function is taken into account, and
optimized features are found from the best solution. Among
179 features, only 83 optimized feature vectors are obtained
using 20 courses of iteration. GA is again applied on the
dataset of size [239 ×8×10] obtained using a Q-IPG
Authorized licensed use limited to: Silicon Institute of Technology. Downloaded on January 25,2024 at 10:19:04 UTC from IEEE Xplore. Restrictions apply.
ACHARYA et al.: OPTIMIZATION TECHNIQUES FOR A VOLTAMMETRIC SIGNAL 19845
TABLE II
PREDICTION OF GAL AN D EC CONCENTRATION IN UNK NOW N SAMPLES
TABLE III
ACT UAL A ND PREDICTED GAL A ND EC CONTENT FRO M TH E LOOCV-BASE D PLSR AND PCR MODEL
electrode. In this case, 129 data points have been optimized
among the 239 data points for the detection of EC in green
tea samples.
2) Data Optimization Using BA:BA is an unsupervised data
optimization algorithm applied to the transformed dataset
of sizes [179 ×8×10] and [239 ×8×10]. In this
work, 20 numbers of bats, which have been optimized by
trial and error methods, are considered for the objective of
the optimization. The optimized feature data 35 and 33 are
obtained among the 179 and 239 feature vectors.
3) Data Optimization Using WOA:WOA has been applied
on the transformed dataset of size [179 ×8×10] and
[239 ×8×10] simultaneously for the detection of GAL and
EC in ten variants of green tea samples. Here, we consider
100 iterations and ten numbers of whales that have been
optimized for the ease of best result on the transformed dataset.
On applying this method, 19 and 14 feature vectors for the
detection of GAL and EC have been optimized. Table Ishows
the comparison between different meta-heuristic optimization
in terms of the number of features obtained after optimization.
The data size reduction may be observed in this table. As the
Authorized licensed use limited to: Silicon Institute of Technology. Downloaded on January 25,2024 at 10:19:04 UTC from IEEE Xplore. Restrictions apply.
19846 IEEE SENSORS JOURNAL, VOL. 23, NO. 17, 1 SEPTEMBER 2023
objective of this work is to build a model which produces
the maximum prediction accuracy rate with minimum feature
vectors, here GA-optimized features have been considered
instead of BA and WOA. It is being observed that BA and
WOA produce a good prediction accuracy rate compared to the
previous work for the detection of GAL and EC with minimum
feature vectors.
B. Calibration and Prediction Analysis
1) PLSR Analysis:The empirical pertinence of the synthe-
sized electrodes has been determined by the PLSR model in
concurrence with the LOOCV method. The DPV response and
HPLC data of GAL and EC were separated into two subsets,
namely, the training and testing set with a ratio of 80:20.
The correlation coefficient (R2)and root mean square error
(RMSE) between the predicted and empirical data have been
used to measure the performance of this model. The LVs 11 for
GAL and 15 for EC have been considered for yielding the
lowest RMSEC values of 0.253 and 0.094 given in Table II.
Here, this model predicts GAL and EC concentration (mg/g)
in green tea with an average prediction accuracy of 92.24%
and 97.95%, which has been depicted in Table III.
2) PCR Analysis:The PCR model customized the relation-
ship between PCs of the data matrix of size [83 ×8×10] and
[129 ×8×10] for GAL and EC with HPLC data. The LVs
14 for GAL and 15 for EC have been considered for yielding
the lowest RMSEC values 0.239 and 0.088. The statistical
metrics for PLSR and PCR regression model have been shown
in Table II, which includes the calibration coefficient(R2
C),
validation coefficient (R2
V), and RMSEC values. The average
prediction accuracies of GAL and EC have been measured to
be 96.24% and 97%, which have been shown in Table III.
C. Statistical Significance of the Result
The average accuracy found using GA is better than BA
and WOA feature selection algorithm, which has been shown
in Table III. Although to check the statistical significance
of the result found in terms of prediction accuracy, the
Mann–Whitney U test (a nonparametric version 2 sample t-
test) has been performed, considering GA with BA and GA
with WOA on prediction accuracy results separately. In both,
the case Mann–Whitney U test compares two populations with
accurate results. The null hypothesis shows that the result for
both cases is identical. Here, the Mann–Whitney U test result
shows the pand zvalues of GA versus BA and GA versus
WOA combinations, and the values are 0.034 (i.e., <0.05) and
2.1169 and 0.018 (i.e., <0.05) and 2.3575, respectively, which
shows the significantly different results. In contrast to BA and
WOA, GA had the lowest standard deviation of the PLSR
and PCR model. In accordance with Table III, the optimum
standard deviation was measured as 6.22 and 6.32 for GAL
and 2.11 and 2.43 for EC. Hence, it may be concluded that
the results obtained by using GA are significantly different
and better than BA and WOA.
V. IMPROVEMENT IN PREDICTION ACC URACY
USI NG T HE PROPOSED MODEL
The literature survey reveals that the prediction accuracy
of the MIP-CuO electrode for the detection of GAL in
green tea samples was 88.97%. This accuracy was achieved
TABLE IV
COMPARISON OF PREVIOUS LITERATURE WITH PRESENT WORK
by deploying electrodes to obtain full dataset, i.e.,179 data
points. The main focus of the work was to enhance prediction
accuracy using optimized reduced features. Table IV shows
the comparison of the work established in previous literature
to that of the present work. Here, a prediction accuracy of
96.24% was achieved by the optimized 83 data points instead
of the full dataset in GAL detection. This result indicates a
prediction model, which increases the prediction accuracy of
the electrochemical system. The validation of the model was
performed on the electrode data obtained in the EC detection
in our previous work. Here, some data transformation and
optimization techniques have been used. This model produced
higher values of prediction accuracy of 97.95% with 129 opti-
mized datasets instead of 94.54% accuracy with 239 datasets
in the previous work. Thus, this model can be deployed
to measure the higher prediction accuracy with an opti-
mized dataset, which decreases the computational time of the
system.
VI. CONCLUSION
The present treatise is focused on the development of a
metaheuristic-based model that increases the overall prediction
accuracy of an electrochemical system by deploying optimized
and reduced datasets than that obtained using raw datasets.
In this work, synthesized electrode responses for the detection
of GAL and EC in green tea samples were transformed by
the DCT technique. Then, the transformed features were opti-
mized independently using three metaheuristic optimization
methods, viz., GA, BA, and WOA to reduce the size of the fea-
ture dataset. The predictability of two synthesized electrodes
MIP-GAL and Q-IPG datasets has been evaluated by PLSR
and PCR models. Herein, GA optimized dataset produced the
maximum prediction accuracy of 96.24% for GAL detection
and 97.95% for EC detection. This significant enhancement
of the prediction accuracy of the model was achieved using
83 and 129 features instead of the full dataset. In summary, this
work proposed a generalized prediction model for the analysis
of voltammetric electrode data for qualitative measurement
of tea and other applications as a generic methodology. The
results so obtained indicate that the generalized prediction
model developed in this work revealed higher prediction
accuracy than that obtained in [17]. To validate the model,
the dataset obtained in [18] has been used, and the same
feature transformation and data optimization techniques have
been deployed.
ACKNOWLEDGMENT
The authors express their gratitude to the Tocklai Tea
Research Institute, Jorhat, Assam, India, for providing the tea
samples and HPLC data.
Authorized licensed use limited to: Silicon Institute of Technology. Downloaded on January 25,2024 at 10:19:04 UTC from IEEE Xplore. Restrictions apply.
ACHARYA et al.: OPTIMIZATION TECHNIQUES FOR A VOLTAMMETRIC SIGNAL 19847
REFERENCES
[1] R. V. Lith and G. A. Ameer, “Antioxidant polymers as biomaterial,” in
Oxidative Stress and Biomaterials, vol. 10. New York, NY, USA: Aca-
demic, 2016, pp. 251–296, doi: 10.1016/B978-0-12-803269-5.00010-3.
[2] P. Carloni et al., “Antioxidant activity of white, green and black tea
obtained from the same tea cultivar,” Food Res. Int., vol. 53, no. 2,
pp. 900–908, Oct. 2013, doi: 10.1016/j.foodres.2012.07.057.
[3] H. G. Valery, A. Chtaini, and B. Loura, “Voltammetric sensor based
on electrodes modified by poly(vinyl alcohol)-natural clay film, for the
detection of gallic acid,” Portugaliae Electrochim. Acta, vol. 37, no. 5,
pp. 327–333, 2019, doi: 10.4152/pea.201905327.
[4] S. Sarafraz, H.-A. Rafiee-Pour, M. Khayatkashani, and A. Ebrahimi,
“Electrochemical determination of gallic acid in Camellia sinensis, Viola
odorata, Commiphora mukul, and Vitex agnus-castus by MWCNTs-
COOH modified CPE,” J. Nanostruct., vol. 9, no. 2, pp. 384–395, 2019,
doi: 10.22052/JNS.2019.02.020.
[5] L. Wang, M. S. Halquist, and D. H. Sweet, “Simultaneous deter-
mination of gallic acid and gentisic acid in organic anion trans-
porter expressing cells by liquid chromatography–tandem mass
spectrometry,” J. Chromatogr. B, vol. 937, pp. 91–96, Oct. 2013, doi:
10.1016/j.jchromb.2013.08.024.
[6] W. Ma et al., “Rapid and specific sensing of gallic acid with a
photoelectrochemical platform based on polyaniline–reduced graphene
oxide–TiO2,” Chem. Commun., vol. 49, no. 71, pp. 7842–7844, 2013,
doi: 10.1039/c3cc43540g.
[7] X. Wang, J. Wang, and N. Yang, “Flow injection chemiluminescent
detection of gallic acid in olive fruits,” Food Chem., vol. 105, no. 1,
pp. 340–345, 2007, doi: 10.1016/j.foodchem.2006.11.061.
[8] K. Dhalwal, V. M. Shinde, Y. S. Biradar, and K. R. Mahadik, “Simulta-
neous quantification of bergenin, catechin, and gallic acid from Bergenia
ciliata and Bergenia ligulata by using thin-layer chromatography,”
J. Food Composition Anal., vol. 21, no. 6, pp. 496–500, Sep. 2008,
doi: 10.1016/j.jfca.2008.02.008.
[9] R. de Queiroz Ferreira and L. A. Avaca, “Electrochemical determination
of the antioxidant capacity: The ceric reducing/antioxidant capac-
ity (CRAC) assay,” Electroanalysis, vol. 20, no. 12, pp. 1323–1329,
Jun. 2008, doi: 10.1002/elan.200704182.
[10] S. Nag et al., “A simple nano cerium oxide modified graphite elec-
trode for electrochemical detection of formaldehyde in mushroom,”
IEEE Sensors J., vol. 21, no. 10, pp. 12019–12026, May 2021, doi:
10.1109/JSEN.2021.3066113.
[11] S. Nag, S. Pradhan, D. Das, B. Tudu, R. Bandyopadhyay, and
R. B. Roy, “Fabrication of a molecular imprinted polyacryloni-
trile engraved graphite electrode for detection of formalin in food
extracts,” IEEE Sensors J., vol. 22, no. 1, pp. 42–49, Jan. 2022, doi:
10.1109/JSEN.2021.3128520.
[12] S. Nag, D. Das, H. Naskar, B. Tudu, R. Bandyopadhyay, and
R. B. Roy, “Detection of metanil yellow adulteration in turmeric
powder using nano nickel cobalt oxide modified graphite electrode,”
IEEE Sensors J., vol. 22, no. 13, pp. 12515–12521, Jul. 2022, doi:
10.1109/JSEN.2022.3178768.
[13] M. Kahl and T. D. Golden, “Electrochemical determination of phenolic
acids at a Zn/Al layered double hydroxide film modified glassy carbon
electrode,” Electroanalysis, vol. 26, no. 8, pp. 1664–1670, Aug. 2014,
doi: 10.1002/elan.201400156.
[14] N. Leibl, K. Haupt, C. Gonzato, and L. Duma, “Molecularly imprinted
polymers for chemical sensing: A tutorial review,” Chemosensors, vol. 9,
no. 6, p. 123, May 2021, doi: 10.3390/chemosensors9060123.
[15] J. W. Lowdon et al., “MIPs for commercial application in low-
cost sensors and assays—An overview of the current status quo,”
Sens. Actuators B, Chem., vol. 325, Dec. 2020, Art. no. 128973, doi:
10.1016/j.snb.2020.128973.
[16] D. Das et al., “Titanium oxide nanocubes embedded molecularly
imprinted polymer-based electrode for selective detection of caffeine in
green tea,” IEEE Sensors J., vol. 20, no. 12, pp. 6240–6247, Jun. 2020,
doi: 10.1109/JSEN.2020.2972773.
[17] D. Das et al., “CuO nanoparticles decorated MIP-based electrode for
sensitive determination of gallic acid in green tea,” IEEE Sensors J.,
vol. 21, no. 5, pp. 5687–5694, Mar. 2021, doi: 10.1109/JSEN.2020.
3036663.
[18] D. Das et al., “Electrochemical detection of epicatechin in green
tea using quercetin-imprinted polymer graphite electrode,” IEEE
Sensors J., vol. 21, no. 23, pp. 26526–26533, Dec. 2021, doi:
10.1109/JSEN.2021.3122145.
[19] M. S. H. Kalathingal, S. Basak, and J. Mitra, “Artificial neural network
modeling and genetic algorithm optimization of process parameters in
fluidized bed drying of green tea leaves,” J. Food Process Eng., vol. 43,
no. 1, pp. 1–7, Jan. 2020, doi: 10.1111/jfpe.13128.
[20] M. Yao et al., “A modified genetic algorithm optimized SVM for rapid
classification of tea leaves using laser-induced breakdown spectroscopy,”
J. Anal. At. Spectrometry, vol. 36, no. 2, pp. 361–367, 2021, doi:
10.1039/D0JA00317D.
[21] G. Ren, Y. Wang, J. Ning, and Z. Zhang, “Highly identification of
keemun black tea rank based on cognitive spectroscopy: Near infrared
spectroscopy combined with feature variable selection,” Spectrochim.
Acta A, Mol. Biomol. Spectrosc., vol. 230, Apr. 2020, Art. no. 118079.
[22] X. Yang and X. He, “Bat algorithm: Literature review and applications,”
Int. J. Bio-Inspired Comput., vol. 5, pp. 141–149, Aug. 2013, doi:
10.1504/IJBIC.2013.055093.
[23] M. Deepak and R. Rustum, “Review of latest advances in nature-inspired
algorithms for optimization of activated sludge processes,” Processes,
vol. 11, no. 1, p. 77, Dec. 2022, doi: 10.3390/pr11010077.
[24] Y.-W. Liu, H. Feng, H.-Y. Li, and L.-L. Li, “An improved whale
algorithm for support vector machine prediction of photovoltaic
power generation,” Symmetry, vol. 13, no. 2, p. 212, Jan. 2021, doi:
10.3390/sym13020212.
[25] M. Mafarja and S. Mirjalili, “Whale optimization approaches for
wrapper feature selection,” Appl. Soft Comput., vol. 62, pp. 441–453,
Jan. 2018, doi: 10.1016/j.asoc.2017.11.006.
[26] S. Acharya et al., “Selection of optimum number of sensors of an
electronic tongue for efficient classification of black tea: A combina-
torial approach based on discrete cosine transform and artificial neural
network,” in Proc. ICRCICN, 2018, pp. 108–111, doi: 10.1109/ICR-
CICN.2018.8718735.
[27] M. Wang, X. Cetó, and M. D. Valle, “A novel electronic tongue
using electropolymerized molecularly imprinted polymers for the
simultaneous determination of active pharmaceutical ingredients,”
Biosensors Bioelectron., vol. 198, Feb. 2022, Art. no. 113807, doi:
10.1016/j.bios.2021.113807.
[28] S. Katoch, S. S. Chauhan, and V. Kumar, “A review on genetic algorithm:
Past, present, and future,” Multimedia Tools Appl., vol. 80, no. 5,
pp. 8091–8126, Feb. 2021, doi: 10.1007/s11042-020-10139-6.
[29] S. Qi, Q. Ouyang, Q. Chen, and J. Zhao, “Real-time monitoring of total
polyphenols content in tea using a developed optical sensors system,”
J. Pharmaceutical Biomed. Anal., vol. 97, pp. 116–122, Aug. 2014, doi:
10.1016/j.jpba.2014.04.034.
[30] M. W. U. Alam, “Improved binary bat algorithm for feature selection,”
M.S. thesis, Dept. Inf. Technol., Åbo Akademi Univ., Turku, Finland,
2019. [Online]. Available: https://www.semanticscholar.org/paper/
Improved-Binary-Bat-Algorithm-for-Feature-Selection-
Alam/1e15e8012125405008e9154e90c02df1cd2ca0ce
[31] R. Ramasamy and S. Rani, “Modified binary bat algorithm for feature
selection in unsupervised learning,” Int. Arab J. Inf. Technol., vol. 15,
no. 6, pp. 1060–1067, 2018.
[32] S. Mirjalili and A. Lewis, “The whale optimization algorithm,”
Adv. Eng. Softw., vol. 95, pp. 51–67, May 2016, doi:
10.1016/j.advengsoft.2016.01.008.
[33] S. Sabzi, R. Pourdarbani, M. H. Rohban, G. García-Mateos, and
J. I. Arribas, “Estimation of nitrogen content in cucumber plant
(Cucumis sativus L.) leaves using hyperspectral imaging data with
neural network and partial least squares regressions,” Chemomet-
ric Intell. Lab. Syst., vol. 217, Oct. 2021, Art. no. 104404, doi:
10.1016/j.chemolab.2021.104404.
[34] Y.-W. Lin, B.-C. Deng, Q.-S. Xu, Y.-H. Yun, and Y.-Z. Liang,
“The equivalence of partial least squares and principal compo-
nent regression in the sufficient dimension reduction framework,”
Chemometric Intell. Lab. Syst., vol. 150, pp. 58–64, Jan. 2016, doi:
10.1016/j.chemolab.2015.11.003.
[35] A. R. Jalalvand, “Engagement of chemometrics and analytical
electrochemistry for clinical purposes: A review,” Chemometric
Intell. Lab. Syst., vol. 227, Aug. 2022, Art. no. 104612, doi:
10.1016/j.chemolab.2022.104612.
[36] O. Valencia, M. C. Ortiz, S. Ruiz, M. S. Sánchez, and L. A. Sarabia,
“Simultaneous class-modelling in chemometrics: A generalization of
partial least squares class modelling for more than two classes
by using error correcting output code matrices,” Chemometric
Intell. Lab. Syst., vol. 227, Aug. 2022, Art. no. 104614, doi:
10.1016/j.chemolab.2022.104614.
Authorized licensed use limited to: Silicon Institute of Technology. Downloaded on January 25,2024 at 10:19:04 UTC from IEEE Xplore. Restrictions apply.