 Access to this fulltext is provided by Springer Nature.
 Learn more
Download available
Content available from Scientific Reports
This content is subject to copyright. Terms and conditions apply.
1
Vol.:(0123456789)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports
AutoML‑ID: automated machine
learning model for intrusion
detection using wireless sensor
network
Abhilash Singh1, J. Amutha2, Jaiprakash Nagar3, Sandeep Sharma4* & Cheng‑Chi Lee5,6*
Momentous increase in the popularity of explainable machine learning models coupled with the
dramatic increase in the use of synthetic data facilitates us to develop a cost‑ecient machine
learning model for fast intrusion detection and prevention at frontier areas using Wireless Sensor
Networks (WSNs). The performance of any explainable machine learning model is driven by its
hyperparameters. Several approaches have been developed and implemented successfully for
optimising or tuning these hyperparameters for skillful predictions. However, the major drawback
of these techniques, including the manual selection of the optimal hyperparameters, is that they
depend highly on the problem and demand application‑specic expertise. In this paper, we introduced
Automated Machine Learning (AutoML) model to automatically select the machine learning model
(among support vector regression, Gaussian process regression, binary decision tree, bagging
ensemble learning, boosting ensemble learning, kernel regression, and linear regression model) and
to automate the hyperparameters optimisation for accurate prediction of numbers of k‑barriers for
fast intrusion detection and prevention using Bayesian optimisation. To do so, we extracted four
synthetic predictors, namely, area of the region, sensing range of the sensor, transmission range of
the sensor, and the number of sensors using Monte Carlo simulation. We used 80% of the datasets
to train the models and the remaining 20% for testing the performance of the trained model. We
found that the Gaussian process regression performs prodigiously and outperforms all the other
considered explainable machine learning models with correlation coecient (R = 1), root mean
square error (RMSE = 0.007), and bias = − 0.006. Further, we also tested the AutoML performance on
a publicly available intrusion dataset, and we observed a similar performance. This study will help
the researchers accurately predict the required number of k‑barriers for fast intrusion detection and
prevention.
Intrusion detection at border areas is of utmost importance and demands a high level of accuracy. Any failure in
intrusion detection may result in havoc on the nation’s security1. Each country shares international boundaries
with its neighboring countries, extending to thousands of kilometers. Continuous monitoring of such a colossal
borderline through occasional patrolling is a crucial problem. To overcome this problem, WSNs are generally
used and deployed along the borderline for surveillance and monitoring2,3. WSNs are a widely adopted technol
ogy that consists of a group of sensors capable of sensing, processing, and transmitting processed information.
It can be easily installed anywhere, even in hardtoreach areas, because it does not require preinstalled infra
structure. e capability of detecting any event or environmental condition makes it more prudent for intrusion
detection applications4,5. Apart from intrusion detection, WSNs found applications in precision agriculture,
health monitoring, environment monitoring, hazards monitoring, and many more6–9.
OPEN
1Indian Institute of Science Education and Research Bhopal, Fluvial Geomorphology and Remote Sensing
Laboratory, Bhopal 462066, India. 2Gautam Buddha University, School of ICT, Greater Noida 201312, India. 3Indian
Institute of Technology Kharagpur, Subir Chowdhury School of Quality and Reliability, Kharagpur 721302,
India. 4Department of Electronics Engineering, Madhav Institute of Technology and Science, Gwalior 474005,
India. 5Department of Library and Information Science, Research and Development, Center for Physical Education,
Health, and Information Technology, Fu Jen Catholic University, New Taipei 242, Taiwan. 6Department of Computer
Science and Information Engineering, Asia University, Taichung 41354, Taiwan. *email: sandeepsvce@gmail.com;
cclee@mail.u.edu.tw
Content courtesy of Springer Nature, terms of use apply. Rights reserved
2
Vol:.(1234567890)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
Border surveillance, intrusion detection, and prevention problems are addressed with two different
approaches. Researchers propose various algorithms and Internet of ings (IoT) solutions for intrusion detec
tion and surveillance in border areas in the rst approach. In the second approach, they develop analytical
models to estimate the intrusion detection probability in terms of kcoverage, kbarrier coverage, number of
kbarriers, and many other performance metrics. Yang etal.10 have proposed an energyecient intrusion detec
tion method that is capable of identifying weak zones of the network deployment region that need to be repaired.
Aer identifying the weak zones, they are repaired to achieve the desired quality of barrier coverage. Specically,
their proposed method focuses on onedirectional coverage only for single and multiple intruder scenarios.
e authors have claimed that their proposed method and algorithms could enhance the network lifetime. In
another work presented in11, Raza etal. have analysed the impact of heterogeneous WSNs deployed following
either uniform or Gaussian distribution scenario. ey have studied the impact of sensor density and sensing
range of sensor nodes on the intrusion detection probability. ey found that the heterogeneous WSNs provide
better intrusion detection performance than the homogeneous WSNs at a given sensing range and sensor node
density. Similarly, Arfaoui etal.12 have rendered an analytical model that considers the notion of possible paths
that an intruder can follow to cross a belt region in border areas. ey have developed a model considering
border area characteristics and the intrusion paths to estimate the time taken by an intruder to cross the border
area. e authors conclude that their proposed model can detect the intrusion as soon as an intruder enters the
restricted border area.
Further, Singh and Singh13 have presented a smart border surveillance system that uses a WSN which is able
to identify and detect the intrusion and then alerts the control center about the presence of an intruder. e
proposed system is capable in dierentiating between animals and persons. Further, the system uses Raspberry
Pi boards integrated with infrared, ultrasonic and camera sensors and is found to be very eective and accurate
to identify any possible intruder. Again, Sharma and Kumar14 have proposed a MLbased smart surveillance
and intrusion detection system for border regions. e proposed system is capable in detection intruders during
day time and at night along with the kind of weapon carried by the intruder. e proposed system is made of a
highresolution camera with IR capabilities for day and night vision, a GPS module interfaced with Raspberry Pi
to extract the accurate location of the intruder, and a bluetooth scanner to detect the bluetooth signature of the
intruder device. e entire module is put into a climate protected box that can be mounted on a high platform.
Further, Mishra etal. in15 have provided a detailed literature review on various ML techniques for intrusion detec
tion. ey have also provided a comprehensive discussion on various types of attacks along with their respec
tive features and security threats. With the help of a specic feature, ML techniques can identify and detect the
intrusion quickly and accurately. Sun etal.16 have proposed a threelevel intrusion detection model to minimise
the memory consumption, computational time, and cost. e proposed model is claimed to decrease memory
consumption, time, and cost up to a great extend. Further, in17, Ghosh etal. have proposed two routing schemes,
namely KPS and LoopFree (LP)KPS, to enhance the lifetime of a WSN deployed for intrusion detection in
border areas or surveillance of some crucial military establishments. On comparing the proposed algorithms
with LEACH and TEEN routing algorithms, they found that the proposed algorithms provide enhanced network
lifetime. In18, Benahmed and Benahmed have proposed an optimal approach to achieve a faulttolerant network
for the surveillance of critical areas using WSNs. e proposed approach identies the faulty sensors and replaces
them with active sensors to ll the coverage gap. e proposed approach can provide a sucient minimum
number of sensors to cover the area under surveillance. Another work presented by Arfaoui and Boudriga in19
provided an ecient surveillance system that can rapidly detect any intruder crossing border areas. In this work,
the authors have incorporated the impact of obstacles present in the environment and the terrain of the border
areas to derive the expression for intrusion detection probability.
Further, Sharma and Nagar20 have obtained an analytical expression of kbarrier coverage probability for
intrusion detection in a rectangular belt region. ey have considered all the possible paths an intruder may
follow to cross the region. Further, they have also analysed the impact of various parameters such as the number
of sensors, sensing range, sensor to intruder velocity ratio, and the intrusion path angle.
e analytical approaches discussed above eectively solve the intrusion detection problem. However, these
approaches need validation through the simulation approach, which is timeconsuming. For example, a single
iteration requires approximately 15 hours for a particular set of network parameters, increasing signicantly
as the network complexity increases. Various machine learning methods have been proposed to overcome the
timecomplexity issue associated with the simulations. Recently, Singh etal.21 proposed three machine learning
methods based on GPR to map the kbarrier coverage probability for accurate and fast intrusion detection using
WSNs. ese methods are based on scaling the predictors; scaleGPR (SGPR), centermeanGPR (CGPR), and
GPR. ey have used synthetic predictors derived from Monte Carlo simulations. ey selected many sensors,
sensing range of the sensor, sensor to intruder velocity ratio, mobile to static node ratio, angle of the intrusion
path, and the required kbarriers as potential predictors. ey found that the nonstandardise methods accurately
map the kbarrier coverage probability using the synthetic variables with R = 0.85 and RMSE = 0.095. More
recently, Singh etal.22 proposed a logarithmic predictor transformation and scalingbased algorithm coupled
with SVR (i.e., LTFSID) to map the number of required kbarriers for fast intrusion detection and prevention
over a rectangular Region of Interest (RoI) considering uniform sensor distribution. e dimension of the dataset
LTFSID is 182
×
5. ey used four predictors to accurately predict the required kbarriers. ey reported that
the proposed approach accurately predicts the kbarriers with R = 0.98 and RMSE = 6.47. e feasibility of deep
learning algorithms for the intrusion detection has been investigated by Otoum etal. in23. ey have presented
a restricted Boltzmann machinebased clustered IDS (RBCIDS) for monitoring critical infrastructures using
WSNs. Further, they have compared the performance of RBCIDS with the adaptively supervised and clustered
hybrid IDS (ASCHIDS) and found that both provides same detection and accuracy rates, but, detection time
of RBCIDS is approximately twice that of ASCHIDS.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
3
Vol.:(0123456789)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
e machine learning methods discussed above involve manual selection of the best performing algorithm,
which may lead to bias results if the results are not compared with the benchmark algorithm. In addition, the
optimisation of the hyperparameter associated with each algorithm is treated dierently. To solve this problem,
in this paper, we introduced an automated machine learning (AutoML) model to automate the model selection
and hyperparameter optimisation task. In doing so, we synthetically extracted potential predictors (i.e., area of
the region, sensing range of the sensor, transmission range of the sensor, and the number of sensors) through
Monte Carlo simulation. We then evaluated the predictor importance and predictor sensitivity through the
regression tree ensemble approach. Subsequently, we applied AutoML on the training datasets to get the best
optimised model. We evaluated the performance of the best performing algorithm over the testing data using
R, RMSE, and bias as performance metrics.
Material and methods
Predictor generation. e quality of the prediction of a machine learning model depends on the quality of
predictors and the model hyperparameters24. ese predictors can be categorised into real and syntheticbased
upon the dataset acquiring process. e real data can be obtained through direct measurements through instru
ments or sensors. However, the generation of real data involves intensive cost and labor. In contrast to real data,
synthetic data can be obtained through mathematical rules, statistical models, and simulations25. In comparison
to real data, acquiring synthetic data is ecient and costeective. Due to this, the use of synthetic datasets to
train machine learning models is increased in the past lustrum21,26–29.
We adopted the synthetic method to extract the predictor datasets using Monte Carlo simulations. In doing
so, we have used network simulator NS2.35 to generate the entire dataset. A nite number of homogeneous (i.e.,
sensing, transmission, and computational capabilities are identical for each sensor) sensor nodes are deployed
according to Gaussian distribution, also known as a normal distribution in a rectangular RoI to achieve this.
Gaussian distribution is considered in this study since it can improve intrusion detection capability and is pre
ferred for realistic applications. In a Gaussian distributed network, the probability that a sensornode is located
at a point (x, y) in reference to the deployed location (x
0
, y
0
)30,31 is given by:
where
σx
and
σy
are the standard deviations of x and y location coordinates, respectively.
To evaluate the performance of WSNs, we have considered the Binary Sensing Model (BSM)32, which is the
most extensively used sensing range model. Each sensor (S
i
) is assumed with the sensing range (R
s
) and is
deployed at an arbitrarypoint (P(x
i
, y
i
)). As per BSM, the target can be detected by any random sensor with
100% probability if the target lies with in the sensing range of the sensor. Otherwise, the target detection prob
ability will be equal to zero and is represented mathematically as:
where
d
(S
i
,P)=
(x
i
−x)
2
+(y
i
−y)
2
, the Euclidean distance between S
i
and target point P. In addition, we
have considered that any two sensors can communicate if they satisfy the criteria, R
tx
≥
2R
s
, where R
tx
and R
s
represents the transmission range and sensing range, respectively. A barrier is constructed by joining a cluster
of sensornodes across the RoI to detect the presence of intruders. Furthermore, to assure barrier coverage, it
is required to identify a Barrier Path (BP) in the RoI. e sensor nodes detect each intruder in the path in this
scenario. us, to ensure guaranteed kbarrier coverage in the rectangular RoI, the number of required nodes is
computed as : k
=⌈ L
2
R
s⌉
and maximum number of BPs can be computed as BP
max
=
⌊N
k⌋
33, where L is the length
of the rectangular RoI, R
s
is the sensing range of nodes, and N is the number of sensor nodes. Table1 lists the
various network parameters and their values that have been used to obtain the simulation results.
Relative predictor importance. In machine learning, the choice of input predictors has a substantial con
trol on its performance28. Predictor importance analysis is not restricted to any particular representations, tech
(1)
f(x,y)=
1
2πσ
x
σ
y
e
−
(x−x0)
2
2σ2
x
+(y−y0)
2
2σ2
y
(2)
P
(Si)=
1, if d(Si,P)≤R
s
0, otherwise
Table 1. Simulation parameters.
Parameters Valu es
Network simulator NS2.35
Network region Rectangular RoI
Network area (
m2
)100
×
50–250
×
200
Sensor nodes (N) 100–400
Sensing range (R
s
) 15–40
m
Transmission range (R
tx
) 30–80
m
Node distribution Gaussian distribution
Sensing model Binary sensing model (BSM)
Content courtesy of Springer Nature, terms of use apply. Rights reserved
4
Vol:.(1234567890)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
niques, or measures and can be used in any situation where predictive models are required. It is used to express
how signicant the predictor was for the model’s predictive performance, irrespective of the structure (linear
or nonlinear) or the direction of the predictor eect. We calculated the relevancy of the selected predictors in
estimating the kbarriers by estimating each predictor’s relative predictor importance score. To do so, we have
used the regression tree ensemble technique21,34. It is an inbuilt class with a treebased classier that assigns a
relative score for every predictor or attribute of the data. e higher the score, the more important the predictor.
Initially, we trained a regression tree ensemble model by boosting hundred regression trees (i.e., t = 100) with
a learning rate of one (i.e.,
δ
= 1) each using the Least Squares gradient Boosting (LSBoost) ensemble aggregation
method. Boosting an ensemble of regression algorithms seems to have several advantages, like, handling missing
data, representing nonlinear patterns, and yielding better generalisation if weak learners were combined into a
single meta learner. In addition, the LSBoost ensemble minimises the mean square error by combining individual
regression trees, oen known as weak learners. e LSBoost technique successfully trains weak learners on the
testing data set, tting residual errors, and detecting its weak points. Based on such weak points, it generates
a new weak learner (
li
) during every iteration. It evaluates its weight (
ωi
) in order to enhance the dierence
between the response value and the aggregated predicted value, hence increasing prediction accuracy. Finally,
the algorithm updates the current model (
Mi
) by emphasising on the prior weak learner’s (
Mi
1) weak point
according to Eq.(3). It then integrates the weak learner into the existing model aer training and iteratively
generates a single strong learner (
Mn
, i.e., ensemble of weak learners).
To explore further the predictor importance, we estimated the coecients indicating the relative importance
of each predictor within the trained model by computing the total variations in the node risk (
R) due to split
among each predictor, and then normalising it by the total number of branch nodes (
RBN
) and is mathemati
cally represented as:
where
RP
indicates the node risk of the parent and
RCH1
&
RCH2
indicates the node risk of two children. e node
risk at individual node (R
i
) is mathematically represented as in Eq.(5);
where
Pi
denotes the probability of node i and
Ei
denotes the node i mean square error.
Predictor sensitivity. We have performed the sensitivity analysis of the predictors using Partial Depend
ence Plot (PDP)21,35. PDP depicts whether a model’s predicted response (outcome) changes as a single explana
tory variable varies. ese plots have the advantage of exhibiting the form of relationship that exists between the
variable and the response36. Moreover, it depicts the marginal eect of one or more variables on the predicted
response of the model37. In this study, we have considered the combined impact of two predictors simultane
ously from the input predictor set (i.e.,
υ
) on the predictand by marginalising the impact of the remaining pre
dictors. To accomplish this, a subset
υs
and a complimentary set (
υc
) of
υs
is extracted from the predictor set
(
υ={z1,z2,...,zn}
) where n represents the total number of predictors. Any prediction on
υ
is determined by
Eq.(6) and the partial dependence of the predictor in
υs
is inferred by computing the expectation (E
c
) of Eq.(6):
where
ρc(υc
) indicates the marginal probability of
υc
, which is represented in Eq.(8).
en, the partial dependency of the predictor in
υs
can be determined by :
where U represents the total number of observations.
Automated machine learning model. AutoML is used to automate the machine learning process
such as data preprocessing, predictor or feature engineering, best algorithm selection, and hyperparameter
optimisation38–40. For past few years, it has been widely used in industry and academia to solve real and near real
time problems41–43. In this study, rstly, we have performed the predictor standardisation using Zscore scaling44.
Aerward, we divided the complete dataset randomly using Mersenne Twister (MT) random generator in an
80:20 ratio for training and testing the AutoML model. e dimension of the complete dataset is 182
×
5, where
(3)
Mi=Mi−1+δ·ωi·li(i=1, 2, 3, ...,n)
(4)
�
R=
R
P
−(R
CH1
+R
CH2
)
RBN
(5)
Ri=Pi·Ei
(6)
f(υ) =f(υ s,υc)
(7)
fs
(υ
s
)=
E
c[
f
(υ
s,
υ
c
)]
=
f(υs,υc)·ρc(υ c)·dυ
c
(8)
ρ
c(υc)≈
p(υs,υc)·dυ
s
(9)
fs(υs)≈
1
U
U
i=1
f(υs,υc
i
)
Content courtesy of Springer Nature, terms of use apply. Rights reserved
5
Vol.:(0123456789)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
182 is the number of observations and 5 is the number of predictors (i.e., area of the region, sensing range of the
sensor, transmission range of the sensor, and the number of sensors) and the response variable (i.e., kbarrier).
e dimension of the training dataset is 145
×
5, and the dimension of the testing dataset is 37
×
5. Aer data
division, we have automated the algorithms selection and hyperparameter optimisation step and investigated
its performance. Various explainable machine learning models participate in the algorithm selection process,
which is discussed next in the upcoming subsections.
Support vector regression model. e Support Vector Regression (SVR) model was introduced by Vapnik
etal.45, and it was developed primarily using the Support Vector Machine (SVM) classiers. e SVR model has
the benet of being able to optimise the nominal margin using regression task analysis and is a popular choice
for prediction and curvetting both for linear and nonlinear regression types46. e relationship among input
and output variables for nonlinear mapping47 is determined by:
where p
=
(
p1
,
p2
,...,
pn)
indicates the input, y
i
∈
Rl indicates the output, w
∈
R
n
indicates the weight vector, q
∈
R indicates the constant, n indicates the number of training datasets and
φ(p)
indicates an irregular function
that is used to assign the input to the predictor. To determine w and q, Eq.(11) is used, where
χi,χ∗
i
indicates
the slack variable.
In the SVR model, the three basic hyperparameters used are the insensitive loss function (
ǫ
) that speci
es the tolerance margin; the capacity parameter or penalty coecient or box constraint (C) that species the
error weight; and the Gaussian width parameter or kernel scale (
γ
)48,49. A high value of C lets SVR reminisce
the training data. e smaller
ǫ
value implies noiseless data. However, the
γ
value is equally responsible for the
underadjustment or overadjustment of prediction. Mathematically, it is represented as:
where K represents the kernel function,
γ
represents the kernel scale that manages the inuence of predictors
variation on kernel variation.
Gaussian process regression model. Gaussian Process Regression (GPR), also known as kriging50 is based on
Bayesian theory51 and is used to solve complex regression problems (high dimension, nonlinearity), facilitates
the hyperparameter adaptive acquisition, easy to implement, and is used with no loss of performance. e
fundamental and extensively used GPR is mainly comprised of a simple zero mean and squared exponential
covariance function52 as represented in Eq. (13).
where
where
k(x,x′)
represents the covariance function or kernels that provide the expected correlation among several
observations. In the GPR model, there are two hyperparameters used, such as the model noise (
̟f
) and the
length scale (g) that regulates the vertical scale and the horizontal scale of the function change, respectively.
Binary decision tree regression. A Binary Decision Tree (BDT) regression is formed by performing consecutive
recursive binary splits on variables, that is of the form y
i
≤
v, y
i
≥
v, where v
∈
R
are observed values in a binary
regression tree53, which is represented as:
where T(y) indicates the regression tree, M indicates the number of tree’s terminal nodes, and B
m
(y) indicates
the base function which is determined by:
(10)
yi=wφ(p)+q
(11)
Minimise :1
2w2 + C
n
i=1
(χi−χ∗
i
)
Subject to
:yi−(wφ(pi)+qi)≤ǫ+χi
(wφ(pi)+qi)−yi≤ǫ+χ∗
i
χi
,
χ
∗
i
≥0
(12)
K(pi
,
p)=e
(−γpi−p
2)
(13)
K
(x,x′)=̟2
fexp
−r
2
(14)
r
=x−x′
2
g
2
(15)
T
(y)=
M
m=1
m·Bm(y
)
(16)
B
m(y)=
L
m
i=1
[yi(m)−vim
]
Content courtesy of Springer Nature, terms of use apply. Rights reserved
6
Vol:.(1234567890)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
where L
m
indicates the total splits, y
i
indicates the involved variable, and v
im
indicates the splitting value. Moreo
ver, the decision tree establishes the rule till the samples in a leaf fall under a specied size, i.e., the minimum
leaf (minleaf) size54. Since the minleaf size denes when splitting must be terminated, it is considered a vital
parameter that must be netuned.
Ensemble regression model. Perrone and Cooper55 proposed a general conceptual framework for obtaining
considerably better regression estimates using ensemble methods. Ensemble Learning (EL) enhances perfor
mance by building and combining several base learners with specic approaches. It is mainly used when there is
a limited amount of training data. It is challenging to choose a suitable classier with this limited available data.
Ensemble algorithms minimise the risk of selecting a poor classier by averaging the votes of individual classi
ers. is study has applied bagging and boosting EL methods due to their widespread usage and eectiveness
for building ensemble learning algorithms.
Bagging (Breiman56,57), also known as bootstrap aggregation or Random Forest (RF), is one of the most promi
nent approach for building ensembles, that uses a bootstrap sampling technique to generate multiple dierent
training sets. Subsequently, the base learners are trained on every training set, and then combining those base
learners to create the nal model. Hence, bagging works for a regression problem as follows: Consider a training
set, S that comprises of data
{(Xi,Yi),i=1, 2, ...,m}
, where X
i
and Y
i
represents the realisation of a multi
dimensional estimator and a real valued variable respectively. A predictor P(YX = x) = f(x)58 is represented as:
At rst, create a bootstrapped sample Eq. (18) based on the empirical distribution of the pairs S
i
= (X
i
, Y
i
),
next, using the plugin concept, estimate the bootstrapped predictor as shown in Eq. (19). Finally, the bagged
estimator is represented by Eq. (20).
Moreover, the three hyperparameters used in bagging are the MinLeafSize (minimum number of observations
per leaf), NumVariablesToSample (number of predictors to sample at every node), and the NumLearningCycles
(number of trees). e rst two parameters determine the tree’s structure, while tuning the nal parameter helps
balance eciency and accuracy.
Boosting (Freund59) is another ensemble method that aims to boost the eciency of a given learning algo
rithm. e LeastSquares Boosting (LSBoost) ensemble method is used in this study because it is suited for
regression and forecasting problems. LSBoost aims to reduce the Mean Squared Error (MSE) between the target
variable (Y) and the weak learners’ aggregated prediction (Y
p
). At rst, median of (Y), represented as
(
Y
) is com
puted. Next, to enhance the model accuracy, several regression trees (r
1
, r
2
,
...
, r
m
) are integrated in a weighted
manner. Individual regression trees are determined by the following predictor variables (X)60:
where (w
m
) represents the weight for the m model, d represents the weak learners, and
η
with
0<η≤
1 repre
sents the learning rate.
Kernel regression model. Kernel regression (Nadaraya61) is the most used nonparametric method on account
of the virtue of kernel and is undoubtedly known as univariate kernel smoother. In order to achieve a kernel
regression, a collection of kernels are locally placed at every observational point. e kernel is set a weight to
every location depending on its distance from the observational point. A multivariate kernel regression62 deter
mines how the response parameter, y
i
is dependent on the explanatory parameter, x
i
, as in Eqs.(22) and(23).
and
where
E[ψi]=Cov[m(xi),ψi]=0
, m(.) represents a nonlinear function, and
ψi
is random with mean zero
and variance
σ2
. It describes the way that y
i
varies around its mean, m(x
i
). e mean can be represented as the
probability density function f:
(17)
ζm(x)=hm(S1,S2,...Sm)(x)
(18)
S∗
i
=(
Y∗
i
,
X∗
i)
(19)
ζ∗
m(x)
=
hm(S∗
1,S∗
2,...S∗
m)(x)
(20)
ζm;B
(
x
)=
P

S∗
m
(
x
)

(21)
Y
p(X)=
Y(X)+η
d
m=1
wm×rm(X
)
(22)
E(yixi)=m(xi)+ψi
(23)
yi=m(xi)+ψi
(24)
m
(xi)=E[Yixi=x]=
y.f(x,y)dy
f(x,y)dy =
y.f(x,y)
dy
f(x)
Content courtesy of Springer Nature, terms of use apply. Rights reserved
7
Vol.:(0123456789)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
Linear regression model. A linear regression model63 examines the relationship among dierent inuential pre
dictors and an outcome variable. e basic linear regression model, which represents the universal set of two
variable and multiple regression as complementary subsets, can be expressed as:
where Y represents the dependent variable,
X1,X2,...,Xn
represents the n independent variables, a and b rep
resents the regression coecients and u represents the stochastic disturbanceterm that could be caused by an
undened independent variable.
Bayesian optimisation. Bayesian Optimisation (BO)64,65 is an ecient approach for addressing optimisation
problems characterised by expensive experiments. It keeps track of the previous observations and forms a proba
bilistic mapping (or model) between the hyperparameter and a probabilistic score on the objective function that
is to be optimised. e probabilistic model is known as a surrogate of the objective function. e surrogate func
tion is much easy to optimise, and with the help of the acquisition function, the next set of hyperparameters is
selected for evaluation on the actual objective function based on its best performance on the surrogate function.
Hence, it comprises a surrogate function for determining the objective function and an acquisition function for
sampling the next observation. In BO, the objective function (f) is obtained from the Gaussian Process (GP) as
described in Eq. (26).
where
µ
and
ϑ
are calculated from the observations of x66.
We select the best performing algorithm among the abovediscussed models with the optimised hyperparam
eter. Lastly, we evaluated the performance of the bestperforming algorithm using the test dataset. A owchart
of the detailed methodology is illustrated in Fig.1.
Results
Predictor importance and sensitivity. We plotted the relative predictor importance score of each pre
dictor along with their respective box plot for a better visual representation of the datasets (Fig.2). We found
that the relative predictor importance score ranges approximately from 9 to 152. e higher the value of the rela
tive estimate, the more relevant is the predictor in estimating the response variable (i.e., kbarriers). We found
that out of these four predictors, the transmission range of the sensor emerges as the most relevant predictor in
predicting the required number of kbarriers for fast intrusion detection and prevention considering Gaussian
node distribution over a rectangular region. e number of sensors also shows good relevancy in predicting the
response variable and ranked second. e area of the region of interest and the sensing range of the sensor shows
fair relevancy and ranked third and fourth, respectively.
We also evaluated the impact of each predictor on the response variable. We plotted the partial depend
ence plot for each possible pair of predictors (Fig.3a–f). For a better visual inspection, we also plotted the
(25)
Y
=a+
n
i=1
biXi+
u
(26)
f(x)∼GP(µ(x),ϑ(xi,xj))
Figure1. Flowchart of the proposed methodology.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
8
Vol:.(1234567890)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
threedimensional plot and its twodimensional illustration. We observed that the area of the RoI has a slightly
negative impact on the target variable i.e., the response variable decreases with an increase in the area of the
RoI. However, an inverse relationship is observed with all other predictors. e sensing range of the sensor, the
transmission range of the sensor, and the number of the sensors have a positive impact on the response variable
i.e., the response variable increases with an increase in these predictors.
Model performance. We iteratively selected the best machine learning model with optimised hyperpa
rameters value using the Bayesian optimisation67–69 on the 80% of the datasets (Fig.4). We used Eq.(27) as the
objective function (Obj) to select the best machine learning model with optimised hyperparameters.
where valLoss is the crossvalidation mean square error (CVMSE). At each iteration, the value of the objective
function is computed for any one of the participating models. e model (with optimised hyperparameters),
which returns the minimum observed loss (i.e., the smallest value of the objective function so far), is considered
as the best model. Aer iterating for 120 iterations, the AutoML algorithm returned the GPR model as the best
model along with the optimal hyperparameters (i.e., for the GPR model; sigma =
0.98
). Before returning the
model, the AutoML algorithm retrains the GPR model on the entire training dataset.
Once we get the trained GPR model, we evaluate its performance on the training datasets to estimate the
training accuracy. We found that the model performed well on the training datasets with a correlation coecient
(R = 1), root mean square error (RMSE = 0.003), and bias = 0. However, for an unbiased evaluation, we evaluated
the performance of the trained model on the test datasets (i.e., 20% of the total datasets). In doing so, we fed the
testing predictors into the trained GPR model and obtained the predicted response. We then compared the GPR
predicted kbarriers with the observed values (Fig.5a). We found that the GRP model performs prodigiously
with a R = 1, RMSE = 0.007, and bias = − 0.006. All the data points are aligned along the regression line and lie
well inside the 95% Condence Interval (C.I).
Further, to assess the appropriateness of the plotted linear regression plot, we performed residual analysis.
We plotted the time series of the observed and the predicted values along with the corresponding residual val
ues (Fig.5b). We found that the residuals are signicantly low and do not follow any pattern, which indicates a
good linear t.
To understand the distribution of the error (i.e., dierence of predicted and observed values), weperformed
error analysis using error histogram (Fig.6). To do so, we plotted the error histogram using ten bins. e error
ranges from
−0.00997
from the le to
0.00356
on the right of the histogram plot. We found that the error follows
a rightskewed Gaussian distribution. e peak of the distribution lies in the underestimated region. Lastly, we
presented the results of the remaining algorithms of the AutoML (i.e., SVR, BDT, Bagging ensemble learning,
Boosting ensemble learning, kernel, and linear regression) in Table2. We found that the best performing AutoML
algorithm (i.e., GPR) outperforms all the other algorithms.
(27)
Obj =log(1+valLoss)
Figure2. Graph showing the relative predictor importance score for all four predictors. e estimates for the
area of the RoI, sensing range of the sensor, transmission range of the sensor, and the number of sensors are
46.0, 9.3, 152.0, and 128.9, respectively.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
9
Vol.:(0123456789)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
Figure3. Twodimensional and threedimensional partial dependency plots show the predictor sensitivity
of all possible predictor pairs. e histogram along the x and yaxis of the twodimensional plot shows the
distribution of the predictor and the response variable, respectively.
Figure4. Curve illustrating the Bayesian optimisation process for the selection of the best machine learning
model with optimal hyperparameters.
Content courtesy of Springer Nature, terms of use apply. Rights reserved
10
Vol:.(1234567890)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
Figure5. e le panel shows the linear regression plot between the predicted and observed responses. e
top plot on the right panel shows the time series plot of the predicted and observed. e bottom panel shows the
corresponding residuals. e dashed line in the residual plot shows the RMSE value.
Figure6. Error analysis using error histogram of 10 bins. e line in red shows the zero error line. e area to
the le of the zero error line shows the underestimated region, and the area right to the zero error line shows the
overestimated region.
Table 2. Performance of the other AutoML algorithms.
Performance metrics
Algorithms
SVR BDT Bagging EL (random
forest) Boosting EL (LSBoost) Kernel regression Linear regression
R 0.93 0.81 0.93 0.73 0.91 0.94
RMSE 63.61 73.07 81.84 118.03 32.29 33.68
Bias 53.59 56.99 67.31 89.81 31.28 31.81
t (s) 95.3 111.3 103.4 107.7 43.01 36.7
Content courtesy of Springer Nature, terms of use apply. Rights reserved
11
Vol.:(0123456789)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
Discussion
We observed that the AutoML approach successfully selects the best machine learning model among a group
of explainable machine learning algorithms (i.e., among SVR, GPR, BDT, bagging ensemble learning, boosting
ensemble learning, kernel regression, and linear regression model) and optimised its hyperparameters. How
ever, we have compared the AutoML derived results with the benchmark algorithms for an unbiased and fair
evaluation of the proposed approach. We selected FeedForward Neural Network (FFNN)70, Recurrent Neural
Network (RNN)71, Radial Basis Neural Networks (RBN)72, Exact RBN73, and Generalised Regression Neural
Network (GRNN)74 as the benchmark algorithms. We selected these algorithms because they are frequently
used in diverse applications such as remote sensing, blockchain, cancer diagnosis, precision medicine, decease
prediction, selfdriving cars, streamow forecasting, and speech recognition; hence have high generalisation
capabilities37,75–77. In doing so, wetrained these algorithms over the same datasets. We found that the AutoML
outperforms all the deep learning benchmark algorithms (Table3). Among the benchmark algorithms GRNN
performs the best (with R = 0.97, RMSE = 64.61, Bias = 60.18, and computational time complexity, t = 2.23 s).
Surprisingly, all the benchmark algorithms have a high positive bias value. It indicates that these models highly
overestimate the number of required kbarriers. We have also compared the performance of the AutoML with
previous studies21,22 for the prediction of kbarriers and kbarrier coverage probability (Table4).
Further, we also tested the performance of the AutoML approach over the publicly available intrusion detec
tion dataset22. In a recent study, Singh etal.22 have proposed a logtransformed feature scaling based algorithm
(i.e., LTFSID) for intrusion detection considering uniform node distribution scenario. We downloaded the
datasets and applied the proposed AutoML approach to them. In doing so, we iterated the AutoML for 120
iterations using the Bayesian optimisation to obtain the best optimised machine learning model. We found that
AutoML approach perform well over the dataset (with R = 0.92, RMSE = 30.59, and Bias = 18.13). Interestingly,
the same GPR algorithms emerges as the best learner algorithms with a optimised sigma = 0.33. It highlights
the potential of the GPR algorithm for intrusion detection, which becomes more apparent from the recently
published literature’s21,78.
e proposed AutoML approach for estimating the kbarriers for fast intrusion detection and prevention
is highly userfriendly and provides a fast solution. It reduces the confusion of selecting the bestperforming
algorithm by automating the process. Further, it also overcomes the limitation of the LTFSID algorithm22.
LTFSID algorithm only works if the input predictors are a positive real number. It will not work if any input
predictors contain zero (or negative values). Although the AutoML approach gives the best result, its perfor
mance will hamper with the sensor aging. In other words, with the aging eect in the sensors, the quality of the
data recorded by the sensor may change drastically (i.e., datasets become dynamic), resulting in performance
degradation. In such a situation, retraining the proposed model will solve the problem.
Conclusion
In this study, we proposed a robust AutoML approach to estimate the accurate number of kbarriers required for
fast intrusion detection and prevention using WSNs over a rectangular RoI considering the Gaussian distribu
tion of the node deployment. We found that the synthetic predictors (i.e., the area of the RoI, sensing range of
the sensornode, transmission range of the sensornode, and the number of sensors) extracted through Monte
Carlo simulations successfully mapped with the kbarriers. Among these predictors, the transmission range
of the sensor emerges as the most relevant predictor, and the sensing range of the sensor emerges as the least
relevant predictor. In addition to this, we observed that only the area of the RoI has a slightly negative impact
on the response variable. We then iteratively run the AutoML algorithms to obtain the best machine learning
model among the explainable machine learning model using Bayesian optimisation techniques. We found that
Table 3. Comparing the performance of the AutoML with the deep learning models.
Performance metrics FFNN RNN Exact RBN RBN GRNN
R 0.47 0.95 0.30 0.41 0.97
RMSE 36.96 14.92 107.95 161.11 64.61
Bias 21.47 71.06 86.21 139.23 60.18
t (s) 2.5 13.51 2.90 3.98 2.23
Table 4. Comparing the results of AutoML with previous studies.
Performance metrics
kbarriers
kbarrier coverage
probability21
AutoML (is study) LTFSID22 GPR SGPR CGPR
R 1 0.98 0.85 0.64 0.79
RMSE 0.007 6.47 0.095 0.137 0.108
t (s) 0.73 0.65 8.16 7.79 9.51
Content courtesy of Springer Nature, terms of use apply. Rights reserved
12
Vol:.(1234567890)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
the AutoML algorithm selects the GPR algorithm as the best machine learning model to map the required
kbarriers accurately. We evaluated the potential of the GPR algorithm over unseen test datasets. We found that
the AutoML elected algorithm performs exceptionally well on the test datasets.
We further compared the AutoML results with the benchmark algorithms for a more reliable and robust
conclusion. We found that AutoML outperforms all the benchmark algorithms in terms of accuracy. For more
generalisation of this approach, we tested the ecacy of the AutoML over the publicly available datasets on
intrusion detection using WSNs, and we found a similar performance. is study is a step towards a costecient
approach for fast intrusion detection and prevention using explainable machine learning models.
Data availability
e datasets generated during and/or analysed during the current study can be made available from the cor
responding author on a reasonable request.
Code availability
e computer algorithms originated during the current study can be made available from the corresponding
author on a reasonable request.
Received: 30 January 2022; Accepted: 18 May 2022
References
1. Chaabouni, N., Mosbah, M., Zemmari, A., Sauvignac, C. & Faruki, P. Network intrusion detection for iot security based on learning
techniques. IEEE Commun. Surv. Tutor. 21, 2671–2701 (2019).
2. Wang, Y., Wang, X., Xie, B., Wang, D. & Agrawal, D. P. Intrusion detection in homogeneous and heterogeneous wireless sensor
networks. IEEE Trans. Mob. Comput. 7, 698–711 (2008).
3. Abduvaliyev, A., Pathan, A.S.K., Zhou, J., Roman, R. & Wong, W.C. On the vital areas of intrusion detection systems in wireless
sensor networks. IEEE Commun. Surv. Tutor. 15, 1223–1237 (2013).
4. Butun, I., Morgera, S. D. & Sankar, R. A survey of intrusion detection systems in wireless sensor networks. IEEE Commun. Surv.
Tutor. 16, 266–282 (2013).
5. Resende, P. A. A. & Drummond, A. C. A survey of random forest based methods for intrusion detection systems. ACM Comput.
Surv. 51, 1–36 (2018).
6. Ali, A., Ming, Y., Chakraborty, S. & Iram, S. A comprehensive survey on realtime applications of wsn. Future Internet 9, 77 (2017).
7. Singh, A., Sharma, S. & Singh, J. Natureinspired algorithms for wireless sensor networks: A comprehensive survey. Comput. Sci.
Rev. 39, 100342 (2021).
8. Amutha, J., Sharma, S. & Nagar, J. Wsn strategies based on sensors, deployment, sensing models, coverage and energy eciency:
Review, approaches and open issues. Wirel. Pers. Commun. 111, 1089–1115 (2020).
9. Nagar, J., Chaturvedi, S. K. & Soh, S. An analytical model to estimate the performance metrics of a nite multihop network deployed
in a rectangular region. J. Netw. Comput. Appl. 149, 102466 (2020).
10. Yang, T., Mu, D., Hu, W. & Zhang, H. Energyecient border intrusion detection using wireless sensors network. EURASIP J.
Wirel. Commun. Netw. 2014, 1–12 (2014).
11. Raza, F., Bashir, S., Tauseef, K. & Shah, S. Optimizing nodes proportion for intrusion detection in uniform and gaussian distributed
heterogeneous wsn. In 2015 12th International Bhurban Conference on Applied Sciences and Technology (IBCAST), 623–628 (IEEE,
2015).
12. Arfaoui, I., Boudriga, N., Trimeche, K. & Abdallah, W. Wsnbased border surveillance systems using estimated known crossing
paths. In Proceedings of the 15th International Conference on Advances in Mobile Computing and Multimedia, 182–190 (2017).
13. Singh, R. & Singh, S. Smart border surveillance system using wireless sensor networks. Int. J. Syst. Assur. Eng. Manage. 20, 1–15
(2021).
14. Sharma, M. & Kumar, C. Machine learningbased smart surveillance and intrusion detection system for national geographic
borders. In Articial Intelligence and Technologies 165–176 (Springer, 2022).
15. Mishra, P., Varadharajan, V., Tupakula, U. & Pilli, E. S. A detailed investigation and analysis of using machine learning techniques
for intrusion detection. IEEE Commun. Surv. Tutor. 21, 686–728 (2018).
16. Sun, Z., Xu, Y., Liang, G. & Zhou, Z. An intrusion detection model for wireless sensor networks with an improved vdetector
algorithm. IEEE Sens. J. 18, 1971–1984 (2017).
17. Ghosh, K., Neogy, S., Das, P. K. & Mehta, M. Intrusion detection at international borders and large military barracks with multi
sink wireless sensor networks: An energy ecient solution. Wirel. Pers. Commun. 98, 1083–1101 (2018).
18. Benahmed, T. & B enahmed, K. Optimal barrier coverage for critical area surveillance using wireless sensor networks. Int. J. Com
mun. Syst. 32, e3955 (2019).
19. Arfaoui, I. & Boudriga, N. A border surveillance system using wsn under various environment characteristics. Int. J. Sens. Netw.
30, 263–278 (2019).
20. Sharma, S. & Nagar, J. Intrusion detection in mobile sensor networks: A case study for dierent intrusion paths. Wirel. Pers. Com
mun. 115, 2569–2589 (2020).
21. Singh, A., Nagar, J., Sharma, S. & Kotiyal, V. A gaussian process regression approach to predict the kbarrier coverage probability
for intrusion detection in wireless sensor networks. Expert Syst. Appl. 172, 114603 (2021).
22. Singh, A., Amutha, J., Nagar, J., Sharma, S. & Lee, C.C. Ltfsid: Logtransformed feature learning and featurescalingbased
machine learning algorithms to predict the kbarriers for intrusion detection using wireless sensor network. Sensorshttps:// doi.
org/ 10. 3390/ s2203 1070 (2022).
23. Otoum, S., Kantarci, B. & Mouah, H. T. On the feasibility of deep learning in sensor network intrusion detection. IEEE Netw.
Lett. 1, 68–71 (2019).
24. Schmidt, J., Marques, M. R., Botti, S. & Marques, M. A. Recent advances and applications of machine learning in solidstate materi
als science. NPJ Comput. Mater. 5, 1–36 (2019).
25. Nikolenko, S.I. etal. Synthetic data for deep learning. arXiv: 1909. 11512 (arXiv preprint) (2019).
26. Chen, R. J., Lu, M. Y., Chen, T. Y., Williamson, D. F. & Mahmood, F. Synthetic data in machine learning for medicine and healthcare.
Nat. Biomed. Eng. 20, 201–5 (2021).
27. R ankin, D. et al. Reliability of supervised machine learning using synthetic data in health care: Model to preserve privacy for data
sharing. JMIR Med. Inform. 8, e18910 (2020).
28. Singh, A., Kotiyal, V., Sharma, S., Nagar, J. & Lee, C.C. A machine learning approach to predict the average localization error with
applications to wireless sensor networks. IEEE Access 8, 208253–208263 (2020).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
13
Vol.:(0123456789)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
29. Abay, N.C., Zhou, Y., Kantarcioglu, M., uraisingham, B. & Sweeney, L. Privacy preserving synthetic data release using deep
learning. In Joint European Conference on Machine Learning and Knowledge Discovery in Databases, 510–526 (Springer, 2018).
30. Wang, D., Xie, B. & Agrawal, D. P. Coverage and lifetime optimization of wireless sensor networks with gaussian distribution. IEEE
Trans. Mob. Comput. 7, 1444–1458 (2008).
31. Wang, Y., Fu, W. & Agrawal, D. P. Gaussian versus uniform distribution for intrusion detection in wireless sensor networks. IEEE
Trans. Parallel Distrib. Syst. 24, 342–355 (2012).
32. Zou, Y. & Chakrabarty, K. Sensor deployment and t arget loca lization in distributed sensor networks. ACM Trans. Embedd. Comput.
Syst. 3, 61–91 (2004).
33. Mostafaei, H., Chowdhury, M. U. & Obaidat, M. S. Border surveillance with wsn systems in a distributed manner. IEEE Syst. J. 12,
3703–3712 (2018).
34. TorresBarrán, A., Alonso, Á. & Dorronsoro, J. R. Regression tree ensembles for wind energy and solar radiation prediction.
Neurocomputing 326, 151–160 (2019).
35. Friedman, J. H. Greedy function approximation: A gradient boosting machine. Ann. Stat. 20, 1189–1232 (2001).
36. Goldstein, A., Kapelner, A., Bleich, J. & Pitkin, E. Peeking inside the black box: Visualizing statistical learning with plots of indi
vidual conditional expectation. J. Comput. Graph. Stat. 24, 44–65 (2015).
37. Singh, A., Gaurav, K., Rai, A. K. & Beg, Z. Machine learning to estimate surface roughness from satellite images. Remote Sens. 13,
3794 (2021).
38. Guyon, I. etal. Design of the 2015 chalearn automl challenge. In 2015 International Joint Conference on Neural Networks (IJCNN),
1–8 (IEEE, 2015).
39. Guyon, I. etal. Automl challenge 2015: Design and rst results. In Proceedings of of AutoML (2015).
40. Guyon, I. etal. A brief review of the chalearn automl challenge: Anytime anydataset learning without human intervention. In
Workshop on Automatic Machine Learning, 21–30 (PMLR, 2016).
41. He, Y. etal. Amc: Automl for model compression and acceleration on mobile devices. In Proceedings of the European Conference
on Computer Vision (ECCV), 784–800 (2018).
42. Feurer, M., Eggensperger, K., Falkner, S., Lindauer, M. & Hutter, F. Practical automated machine learning for the automl challenge
2018. In International Workshop on Automatic Machine Learning at ICML, 1189–1232 (2018).
43. He, X., Zhao, K. & Chu, X. Automl: A survey of the stateoftheart. Knowl.Based Syst. 212, 106622 (2021).
44. Townes, F. W., Hicks, S. C., Aryee, M. J. & Irizarry, R. A. Feature selection and dimension reduction for singlecell rnaseq based
on a multinomial model. Genome Biol. 20, 1–16 (2019).
45. Vapnik, V. et al. Support vector method for function approximation, regression estimation, and signal processing. Adv. Neural Inf.
Process. Syst. 20, 281–287 (1997).
46. Saha, A. et al. Flood susceptibility assessment using novel ensemble of hyperpipes and support vector regression algorithms. Wate r
13, 241 (2021).
47. Arifuzzaman, M., Aniq Gul, M., Khan, K. & Hossain, S. Application of articial intelligence (ai) for sustainable highway and road
system. Symmetry 13, 60 (2021).
48. da Silva Santos, C. E., dos Santos Coelho, L. & Llanos, C. H. Nature inspired optimization tools for svmsniots. MethodsX 8, 101574
(2021).
49. Zaghloul, M. S., Hamza, R. A., Iorhemen, O. T. & Tay, J. H. Comparison of adaptive neurofuzzy inference systems (ans) and
support vector regression (svr) for datadriven modelling of aerobic granular sludge reactors. J. Environ. Chem. Eng. 8, 103742
(2020).
50. César de Sá, N., Baratchi, M., Hauser, L. T. & van Bodegom, P. Exploring the impact of noise on hybrid inversion of prosail rtm
on sentinel2 data. Remote Sens. 13, 648 (2021).
51. Rasmussen, C. E. Gaussian processes in machine learning. In Summer School on Machine Learning 63–71 (Springer, 2003).
52. AsanteOkyere, S., Shen, C., Yevenyo Ziggah, Y., Moses Rulegeya, M. & Zhu, X. Investigating the predictive performance of gauss
ian process regression in evaluating reservoir porosity and permeability. Energies 11, 3261 (2018).
53. Artime Ríos, E. M., Sánchez Lasheras, F., Suárez Sánchez, A., IglesiasRodríguez, F. J. & Seguí Crespo, M. D. M. Prediction of
computer vision syndrome in health personnel by means of genetic algorithms and binary regression trees. Sensors 19, 2800 (2019).
54. Kim, S.H., Moon, I.J., Won, S.H., Kang, H.W. & Kang, S. K. Decisiontreebased classication of lifetime maximum intensity
of tropical cyclones in the tropical western north pacic. Atmosphere 12, 802 (2021).
55. Perrone, M. P. & Cooper, L. N. When networks disagree: Ensemble methods for hybrid neural networks. Tech. Rep., Brown Univ
Providence RI Inst for Brain and Neural Systems (1992).
56. Breiman, L. Bagging Predictors (Technical Report 421) (University of California, 1994).
57. Breiman, L. Stacked regressions. Mach. Learn. 24, 49–64 (1996).
58. Erdal, H. & Karahanoğlu, İ. Bagging ensemble models for bank protability: An emprical research on Turkish development and
investment banks. Appl. So Comput. 49, 861–867 (2016).
59. Freund, Y. et al. Experiments with a new boosting algorithm. In icml Vol. 96 148–156 (Citeseer, 1996).
60. Jung, C. High spatial resolution simulation of annual wind energy yield using nearsurface wind speed time series. Energies 9, 344
(2016).
61. Watson, G. S. Smooth regression analysis. Sankhyā Indian J. Stat. Ser. A 20, 359–372 (1964).
62. Heo, G.Y. Condition monitoring using empirical models: Technical review and prospects for nuclear applications. Nucl. Eng.
Tec hnol. 40, 49–68 (2008).
63. Poole, M. A. & O’Farrell, P. N. e assumptions of the linear regression model. Trans. Inst. Brit. Geograph. 20, 145–158 (1971).
64. Močkus, J. On bayesian methods for seeking the extremum. In Optimization Techniques IFIP Technical Conference 400–404
(Springer, 1975).
65. Feurer, M. etal. Methods for improving Bayesian optimization for automl. In Proceedings of the International Conference on Machine
Learning (2015).
66. Savaia, G. et al. Experimental automatic calibration of a semiactive suspension controller via Bayesian optimization. Control. Eng.
Pract. 112, 104826 (2021).
67. Pelikan, M., Goldberg, D.E., CantúPaz, E. etal. Boa: e Bayesian optimization algorithm. In Proceedings of the Genetic and
Evolutionary Computation Conference GECCO99, vol.1, 525–532 (Citeseer, 1999).
68. Shahriari, B., Swersky, K., Wang, Z., Adams, R. P. & De Freitas, N. Taking the human out of the loop: A review of Bayesian opti
mization. Proc. IEEE 104, 148–175 (2015).
69. Frazier, P.I. A tutorial on Bayesian optimization. arXiv: 1807. 02811 (arXiv preprint) (2018).
70. Fine, T. L. Feedforward Neural Network Methodology (Springer Science & Business Media, 2006).
71. Zaremba, W., Sutskever, I. & Vinyals, O. Recurrent neural network regularization. arXiv: 1409. 2329 (arXiv preprint) (2014).
72. Karayiannis, N. B. Reformulated radial basis neural networks trained by gradient descent. IEEE Trans. Neural Netw. 10, 657–671
(1999).
73. Çivicioğlu, P., Alçı, M. & Bedok, E. Using an exact radial basis function articial neural network for impulsive noise suppression
from highly distorted image databases. In International Conference on Advances in Information Systems, 383–391 (Springer, 2004).
74. Specht, D. F. et al. A general regression neural network. IEEE Trans. Neural Netw. 2, 568–576 (1991).
75. Guidotti, R. et al. A survey of methods for explaining black box models. ACM Comput. Surv. 51, 1–42 (2018).
Content courtesy of Springer Nature, terms of use apply. Rights reserved
14
Vol:.(1234567890)
Scientic Reports  (2022) 12:9074  https://doi.org/10.1038/s4159802213061z
www.nature.com/scientificreports/
76. Xie, M., Li, H. & Zhao, Y. Blockchain nancial investment based on deep learning network algorithm. J. Comput. Appl. Math. 372,
112723 (2020).
77. Shrestha, A. & Mahmood, A. Review of deep learning algorithms and architectures. IEEE Access 7, 53040–53065 (2019).
78. Nwakanma, C.I., Ahakonye, L. A.C., Lee, J.M. & Kim, D.S. Selecting gaussian process regression kernels for iot intr usion detec
tion and classication. In 2021 International Conference on Information and Communication Technology Convergence (ICTC),
462–465 (IEEE, 2021).
Acknowledgements
We want to acknowledge IISER Bhopal, Madhya Pradesh, India; Gautam Buddha University, Uttar Pradesh,
India; IIT Kharagpur, West Bengal, India; MITS Gwalior, Madhya Pradesh, India; Fu Jen Catholic University,
Taiwan; and Asia University, Taiwan, for providing institutional support.
Author contributions
A.S. developed the models, J.N. and J.A. extracted the datasets, S.S. and C.C.L. analysed the results. All the authors
contributed to the writing and reviewed the manuscript.
Competing interests
e authors declare no competing interests.
Additional information
Correspondence and requests for materials should be addressed to S.S.orC.C.L.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional aliations.
Open Access is article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. e images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.
© e Author(s) 2022
Content courtesy of Springer Nature, terms of use apply. Rights reserved
1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small
scale personal, noncommercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be noncommercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commonslicensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal noncommercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for interlibrary loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com
Content uploaded by Abhilash Singh
Author content
All content in this area was uploaded by Abhilash Singh on May 31, 2022
Content may be subject to copyright.