Journal of Ambient Intelligence and Humanized Computing

https://doi.org/10.1007/s12652-020-02476-z

ORIGINAL RESEARCH

An efficient parameter optimization of software reliability growth

model by using chaotic grey wolf optimization algorithm
P. Dhavakumar1 · N. P. Gopalan2

Received: 14 October 2019 / Accepted: 14 August 2020

© Springer-Verlag GmbH Germany, part of Springer Nature 2020

Abstract
Software reliability growth model (SRGM) with modified testing-effort function (TEF) is a function to evaluate and foresee
the parameters of the data. Reliability of software is portrayed as the distinct possibility that for a predefined time, a soft-
ware package will continue to run on an advance domain without frustration. SRGM utilized a few optimization procedure
algorithms to advance the parameters by bifurcating them into a few stages however to upgrade the technique by using all
of the parameters at the same time, the algorithm utilized is the chaotic grey wolf optimization algorithm (CGWO). CGWO
is an advanced heuristic system for portraying the execution by achieving complex parameter optimization and designing
application issues. Different parametric reliabilities rely upon the attributes or characteristics of the data. The parameters
are predicted using the Pham–Zhang (PZ) model. Tandem computer software dataset DS1 and DS2 are used to compare the
predicted parameter of SRGM obtained by Pham–Zhang (PZ) model using testing effort functions (TEFs) based on the evalu-
ation metrics mean square error (MSE), relative error (RE) and coefficient of determination (­ R2). To enhance the reliability
of SRGM, the parameters of SRGM estimated using TEF and enhanced using chaotic maps to improve search performance.
By using the constrained benchmark functions the results of chaotic maps are obtained. Based on the chaotic graph results,
the Chebyshev graph shows a good convergence rate of 78%. Overall, 86% of the results revealed an association between the
choice variable and fitness criteria for CGWO. In the SRGM using CGWO, the expected result is completely mechanized
and does not require any client necessity.

Keywords Software reliability growth model · Testing effort functions · Chaotic grey wolf optimization algorithm · Mean
square error · Relative error · Coefficient of determination

1 Introduction the software quality by perceiving imperfection inclined mod-

Software reliability acting as a critical function in detecting
the software defects that leads to a fiasco in the software sys-
tem (Cheng et al. 2016; Jin and Jin 2016). Software defect
prediction (SDP) allows the resource to be allocated fairly by
predicting the largest number of defective modules accurately
(Wu et al. 2018). Software defect normally delivers wrong or
unforeseen outcomes and practices in unintended ways. Utiliz-
ing deformity indicators can lessen the expenses and improve
* P. Dhavakumar
dhavakumar@pmu.edu
1
Department of Computer Science and Engineering, Periyar
Maniammai Institute of Science and Technology, Thanjavur,
Tamil Nadu, India
2
Department of Computer Applications, National Institute
of Technology, Trichy, India
ules (occurrences) preceding testing such that software engi-
neers can adequately upgrade the designation of constrained
resources for testing and support (Li et al. 2018). The challeng-
ing task in software reliability is the absence of understanding
the software nature when estimating the software (Mallikhar-
juna and Anuradha 2015). In the software lifecycle the early
stage is to find the precision of software reliability prediction by
extrapolating the conduct of the past controller software deliv-
erance (Vizarreta et al. 2018). Software reliability prediction
is ended based on the necessities and the feedbacks time-hon-
oured from the client thrust of view but this progression takes
time to grasp the response from the client (Jin and Jin 2016).
The ideal approach to quantify the reliability and different per-
spectives of software is preposterous to decide (Mallikharjuna
and Anuradha 2015). The software developers made use of the
software reliability model instead of retrieving the feedback
from the client (Jin and Jin 2016).

13
Vol.:(0123456789)

The software reliability paradigm has two approaches.
The first is a software reliability prediction strategy by sche-
matically constrained and the second is to predict software
reliability by testing data (Jin and Jin 2016). SRGM analyze
the performance and reliability of software application and
overlook the internal structure of the application (Singh et al.
2015). It is being exhaustively linked to SRGM to quan-
tify important Software Reliability Measuring like interims
between successive failure, failure intensity, software avail-
ability, maintenance and estimating the mathematical frame-
work (Zeephongsekul et al. 2016). Exponential functions,
non-homogenous Poisson process (NHPP) function, etc.
in combination with time utilizing the fundamental arsenal
of scientific functions factually maps the peaked conduct
of programming in SRGM and incorporate imperfect fault
removal (Cascone et al. 2008; Cutolo et al. 2012; D’Apice
et al. 2011; Rarità et al. 2010). An efficient technique is to
estimate the parameters using the expectation conditional
maximization (ECM) algorithm applied to Non-homogene-
ous NHPP SRGM developed by Zeephongsekul et al. (2016).
In software failure data, time domain and time interval func-
tions were illustrated using derivatives from the theory of
stochastic point processes. They made similar comments on
the ECM algorithm of the Gamma and Weibull failure mod-
els. They clearly demonstrated that three parameters could
be calibrated with this strategy and maximum likelihood
(MLE) estimations from failure model could be discovered.
The foreseeable future failure sort of behaviour can still be
better mapped in a scientific function which can illustrate
past failure with time apparently (Jin and Jin 2016).
The ideal characteristic of software reliability is the Fault
reduction factor (FRF) by processing the sum of all removed
faults in extent to the accomplished failures and failure
detection rate using numeric metrics (Abd-Elkader et al.
2018; Kim et al. 2015; Lakshmanan and Ramasamy 2015;
Fera et al. 2018; Packianather et al. 2014). The relation-
ship between software metrics and the number of faults are
described by learned model (Yang et al. 2015). SRGMs have
proposed various sorts of software reliability growth models
to predict the relationship between software program failure
and software program testing time. To recognize whether
the software reliability is extending sufficiently in fulfilling
the suitable software essential specification, SRGMs uses
software failure time data or the testing results (Kim et al.
2015; Yang et al. 2015; Roy et al. 2017). Fault failure rela-
tionship is determined by FRF (Abd-Elkader et al. 2018).
When accessing the reliability of the framework, the focus is
to check whether the parameter increment or diminishes by
determining the utilization of various parameters (Diwaker
et al. 2018). In SRGM, there are some limitations in optimiz-
ing the parameters in the software framework while part-
ing them into different groups. By using the SRGM with
TEF, the solution for the problem is destined and all the
13
P. Dhavakumar, N. P. Gopalan
parameters are optimized simultaneously in the proposed
technique (Jin and Jin 2016).
This paper proposes, by utilizing the Grey wolf optimiza-
tion algorithm globally optimizing and predicting the param-
eters efficiently, chaos have been introduced for developing
chaotic grey wolf optimization algorithm. When compared
with CGWO, the GWO manages to discover the worldwide
most favourable solution. Instead of the critical parameters
utilized in GWO, different unidirectional chaotic maps are
explored. Chaotic maps based on the nature of reliability, it
assesses in the field of optimizing algorithms by scrutiniz-
ing the search space more efficiently. The results revealed,
that, by comparing with other algorithms and applications,
CGWO can conspicuously overcome conventional GWOs
with an excellent convergence (Kohli and Arora 2018;
Decoderz 2019).
During several research projects the reliability of the
new techniques being used in the health systems, computer
engineering, software engineering and mechanical engi-
neering has already been forecasted and predicted. These
paradigms can be used to forecast reliability, to use software
and hardware and are referenced by the anonymous authors:
Researchers have predicted accuracy in estimating software
reliability by using testing process factors like testing-effort
(TE) and imperfect debugging (ID) by Li et al. (2015). They
improved the dependability stage and the development pro-
cedure of the software items. In the testing phase, they esti-
mated and predicted the failures of data collected by using
SRGMs. The results showed that Inflected S-shaped NHPP
SRGMs performed better prediction by comparing delayed
S-shaped NHPP SRGM.
The implementation of QPSO for streamlining SRGM
parameters with S-shaped TEF is being examined by Jin
and Jin (2016). In their work they have showed minimum
prediction error and using Inflected S-shaped model (ISSM)
the selected datasets were accurate than TEF. The results
showed that the predicting performance of real failure data
was comparatively high based on the traditional methods.
Efficient technique to estimate the parameters using the
expectation conditional maximization (ECM) algorithm
applied to Non-homogeneous poisson process (NHPP)
SRGM developed by Zeephongsekul et al. (2016). In soft-
ware failure data, time domain and time interval functions
were illustrated using derivatives from the theory of sto-
chastic point processes. They made similar comments on the
ECM algorithm of the Gamma and Weibull failure models.
They clearly demonstrated that three parameters could be
calibrated with this strategy and maximum likelihood (MLE)
estimations from failure model could be discovered.
SRGM parameter estimation using Harmony Search
variant (HS) has been instituted by Choudhary et al. (2017).
They conducted various experiments with datasets and by
comparing with least square estimation (LSE) and cuckoo
An efficient parameter optimization of software reliability growth model by using chaotic…
search (CS) approaches, the results showed that the param-
eters were estimated efficiently. A statistical (F-test) analysis
revealed HS superior to the other two (LSE and CS) strat-
egies. Roy et al. (2017) proposed the strategy to enhance
software reliability performance based on asynchronously
weighted dynamic combination platform based on the arti-
ficial neural network (ANN). The experiments have shown
effective evaluation of three real failure data sets. The par-
ticle swarm optimization neural network is compared with
conventional PSO neural network and proved that PSO pro-
duced better results for predicting software reliability accu-
rately and efficiently.
PSO and fuzzy logic were the soft computing techniques
used in researches for estimating and predicting reliability
by Diwaker et al. (2018). In the genetics, radiology, physiol-
ogy, cardiology and neurology fields, medical researchers
have used numerous soft computing strategies. Factors such
as component interaction, dependence on the components,
complexity, failure rate and reusability have been the real
challenge for CBSR component-based software Reliabil-
ity prediction, and the inference that CBSE (Component-
based software Engineering) is the right technology, with
less efforts and the ability to accomplish considerably better
parameter prediction. Riaz et al. (2018) has proposed a new
method of Easy Ensemble Rough-KNN noise filtering to
handle unbalanced software failure prediction classes. By
using this strategy they have eliminated a noisy sample and
resolved the problem. By eliminating irrelevant, repetitive
functions, they have augmented the quality of software data.
The remaining portion of the paper is structured as
defies: The strategy instituted for the parameter estima-
tion of SRGM, the definition of TEFs, GWO and CGWO is
emphasised in Sect. 2. Section 3 specifies the outcomes of
our proposed technique and Sect. 4 ends the strategy pro-
posed and summarizes the future scope.
2 Proposed approach
2.1 Parameter estimation of SRGM
The literature classifies almost all software reliability growth
models to approximate SRGM parameters. This paper exam-
ines the exact measurement of the reliability of software and
how a growth model is consistently used to predict reliability
through the chaotic technique of wolf optimization. When
the vulnerabilities detected are rectified over duration of
time, software reliability growth can be ascertained. There
are various traditional software reliability growth models;
they are Goel–Okumoro (GO) model and Yamada delayed
S-shaped (YDSS) model uses two parameters. Generalized
Goel (GG) model and Inflected S-shaped (INFS) model
uses three parameters. Pham–Zhang (PZ) model uses five
parameters (Zhu and Pham 2018). In the chaotic grey wolf
optimization algorithm five parameters are used so we use
the Pham–Zhang model to predict the parameters. Virtually
each SRGM implies a called function to quantify the number
of errors within a certain time t. The scheme proposed uses
five different TEFs for comparing the SRGM parameters and
detecting the best fit model using the metrics mean square
error, relative error and coefficient of determination for opti-
mizing the parameter in CGWO. We portray sporadically in
the following paragraph a well-regarded SRGM model that
we use in our recent study.
Pham–Zhang (PZ) model: It predicts more accurately
than any other SRGM model, the reliability model that best
describes the collected failure data. At a desired time t, the
mean value function describes the expected number of soft-
ware errors, where the mean value function is m(t). In this
model, the introduction rate is an exponential function of
testing time and the error deduction rate is non-decreasing
with an inflected S-shaped model:
1 a
m(t) = [(z + x)(1 − e−yt )] − (e−𝛽t − e−yt ).
1 + 𝛼e−yt b−𝛽
(1)
Testing effort functions (TEFs): By using TEF function
the prediction process is described precisely. Various TEF
functions used in this approach are Weibull, exponential
Weibull, logistic, generalized logistic and inflexion S-shaped
so that the parameters are made reliable and the prediction
of parameters performed accurately using MSE for DS1 and
DS2. The Weibull TEF can be expressed as,
Sw (t) = N(1 - exp[ - btm ]). (2)
The total amount of consumed testing effort is N, b is
the scale and m is the shape parameter, it performs better
when m > 3.
The exponential Weibull TEF can be given as,
Sw (t) = N(1 - exp[ - btm ])𝜃 (3)
here N is the total amount of consumed testing effort, b is
the scale, m is the shape parameter, it performs better when
m > 3 and θ equals 1. The logistic TEF can be given as,
N
Sl (t) = (4)
1 + A exp[1 - 𝛼t]
where A is a constant and α is the testing-effort consumption
rate. The generalized logistic TEF can be given as,
( )
K + 1∕𝛽
Sgl (t) = N (5)
1 + A exp[−𝛼kt]
where k is a structuring index and β is a constant (Yang et al.
2015; Mirjalili et al. 2014). The inflexion S-shaped TEF can
be expressed as,
13
 P. Dhavakumar, N. P. Gopalan

(1 - exp[ - 𝛼t]) Gray wolves isolate prey during hunting as per the
Sis (t) = N (6) established rules during the encircle prey protocol (as
(1 + Aexp[𝛼t])
shown in Fig. 1):
where α represented as the consumption rate and A is a con-
stant parameter.
⇀ |⇀ ⇀ ⇀ |
P = ||R ⋅ Y p (t) − Y(t)|| (7)
| |
2.2 Methodology
⇀ ⇀ ⇀ ⇀
Y(t + 1) = Y p (t) − V ⋅ P. (8)
Grey wolf optimization (GWO) is an optimization heuristic
based on the gray wolves selection criteria, was developed
by Mirjalili et al. (2014). This is a meta-heuristic algorithm, In Eqs. (7) and (8), where t is labelled iteration number.
strengthened by the apparent framework of gray wolves’
⇀
Y P and ⇀Y⇀ are labelled ⇀level of prey and grey wolf respec-
hunting actions and governance activity. In GWO, global tively. V = 2⃗a.⃗r1 − a⃗ , R = 2 r 2 , are depend on the iteration
⇀

search cannot be carried out in every search and then it and random vectors [0, 1], the characteristics of the param-
becomes critical to discover the globally optimal solution eter a are reduced consecutively from 2 to 0, where r 1
⇀ ⇀

in certain instances. The search functionality was accom- and r 2 are random vectors. Hunting the prey procedure is
⇀

plished on a continual basis. The GWO procedure used for performed under the preceding regulation,
searching the prey using optimization technique is hierarchi-
|⇀ ⇀ |
cal procedure, encircle prey, hunting, attacking prey and the
⇀ ⇀
Pi = ||Ri ⋅ Y i (t) − Y(t),|| (9)
search for prey. In the hierarchical procedure, the wolves are | |
grouped into four categories.
where i denotes α, β, δ:
α, β and δ are the first three leader wolves for controlling
and guiding the other wolves’ ω in the hunting process. In ⇀ ⇀ ⇀ ⇀
Y(t + 1) = 𝛴i={𝛼,𝛽,𝛿} Y i (t) − V i ⋅ Pi . (10)
encircle prey, grey wolves surround prey during the hunting
process by indicating the position of the prey. After recog- In Eqs. (9) and (10), where Xi denote position of leader
nition of prey position, hunting procedure performed by ω
⇀ ⇀
wolves, Ri and V i are random vector. Assaulting and find-
wolves in the guidance of leader wolves. The position of
⇀
ing prey are denoted based on the value of vector V , R⇀
. As
ω wolves around the prey is updated to the guiding leader instantiated in Fig. 1, analyse convened when A is greater
wolves so that the position of prey is estimated. The attack- than 1 or A is less than − 1, otherwise C is greater than 1.
ing prey procedure performs exploitation process. Search In contrast, the misappropriation is convened when |V| < 1
for prey performs exploration process and jumps out from and |R| < 1 (Yazdanbakhsh et al. 2016).
the optimal solution.

Fig. 1  Grey wolf optimization

mechanism

13
An efficient parameter optimization of software reliability growth model by using chaotic…
To improve the searching behaviour of the GWO a new,
efficacious approach to augment SRGM metrics is instituted
in this paper. An enhanced grey wolf optimization algorithm
with an interoperable chaotic search strategy for systemati-
cally strengthening the searching and decrease the chances
of optimized prediction via the CGWO algorithm. Various
kinds of chaotic maps (Appendix Table 4) are being uti-
lized to establish a chaotic variable to generate a chaotic
algorithm.
When considering the chaotic map, the chaotic search
approach is proposed and defined using:
Cxin+1 = 𝜇 ⋅ Cxin (1 − Cxin ).
In Eq. (11), where Cxin denotes the chaotic variable and n
denotes the number of iterations. These chaotic maps have
been widely used by researchers, mathematicians and in
medical science in the optimization field. It obviously helps
to efficiently traverse the search. The number [0, 1] is the ini-
tial value of chaotic maps. The original value of these kinds
of maps on the statistical basis of the literature is 0.7. Each
chaotic map deviates in type of attitude. The chaotic maps
themselves aid to determine the data. The fitness of all user
defined gray wolves in the target area is examined and clas-
sified yet according to their situation with various standard
benchmarking features. It consist of two functions, they are
given as objective function and constraining infringement
function, formulate all the minimal problems. The intent of
the functionality x is to determine the best solution in the
stipulated search space. It can be articulated as:
minQ(y), y = (y1 , y2 , y3 , … yn ) ∈ Pn .
In Eq. (12), where n estimated number of configurations
of a viable solution Y ∈ Q ∈ S , Q is the possible region in
the search space S which defies a P is an n-dimensional rec-
tangle. Thus the rectangle P has domains size in the form
of lower bound and upper bound, lower bound is denoted as
(l) and the upper bound is denoted as (u) as articulated in
Eq. (13) and Q space is the form of Eq. (14) that the number
of limitation say g(g > 0) are described in the Q space.
l(j) ≤ y(j) ≤ u(j), 1≤j≤n (13)
rk (y) ≤ 0, for k = 1, … , v,
(14)
sk (y) = 0, for k = v + 1, … , g.
If any solution describes x satisfies the restriction ru or
su in Q space, then ru is articulated to be an active restric-
(11) tion at y (Lakshmanan and Ramasamy 2015) in Eq. (14),
rk (y) and sk (y) are considered as inequality and equal-
ity restriction respectively. The proposed chaotic grey
wolf optimization algorithmic rule is utilized (as shown
in Algorithm 1) for solving optimization problems effi-
ciently. In the algorithm, the first step involves starting of
folk of grey wolves. Then the chaotic map value is started
⇀
continuously as x0 . Sequentially, the parameter a⃗ , A ⇀
,C
are involved in performing exploration and exploitation
process. Then parameter t denotes the iterations. Fitness
⇀ ⇀ ⇀
of each search wolf is evaluated based on X 𝛼 , X 𝛽 and X 𝛿 .
The first wolf is denoted by 𝛼 , second wolf denoted by 𝛽
and third wolf denoted by 𝛿 . Based on the iterative value
the grey wolves are classified according to their fitness.
The chaotic number is upgraded with the chaotic map
equation. According to Eq. (11), for each search wolf the
position is updated and the parameter value gets updated.
Then replacement of worst fit wolf with best fit wolf is
performed. The fitness of 𝛼 wolf at the later part of the
(12)
iteration is seen in the CGWO algorithm as the optimal
solution. This approach produces better results and save
computing time productively (Yazdanbakhsh et al. 2016;
Yu et al. 2017).
13

2.3 Evaluation metrics
2.3.1 Mean squared error (MSE)
The model prediction of Mean squared error approach
can denote the contrast between genuine perception and
perception esteem. By not hurting the models predictive
capacity, the utilization of data fits the model to a certain
degree and by not expelling the required variable. MSE
formulated as:
13
P. Dhavakumar, N. P. Gopalan
∑n
MSE = (qk − q�k )2 (15)
k=1
where [in Eq. (15)] qk describes the total number of detected
faults at execution time tk based on the actual data, q′k is the
predicted total number of detected discrepancies by time tk
and number of perceptions in the software failure dataset.
Lesser the value of MSE denotes minimum fitness error
and better Software reliability growth model performance
(Jin and Jin 2016) During the testing phase, SRGM con-
sumes huge amount of resource like manpower and CPU
An efficient parameter optimization of software reliability growth model by using chaotic…

hours that is inconsistently allocated during testing. To dis- ∑k

i=1
[M(ti ) − N(ti )]2
crete the resources it is described by utilizing the Testing- 2
R =1− ∑k . (17)
effort functions. Parameters in various kinds of TEF are i=1
[N(ti − N)]2
articulated by MSE. Some criterion is used to make com-
In Eq. (17), where M(ti ) is the estimated number of fail-
parison based on the performance of different TEFs in order
ures and N(ti ) is the observed number of failures at time
to select the best fitting TEF. The five most popular models
(ti ), for (i = 1, 2, 3, … k). The entire number of observation
are used for comparison using DS1 and DS2 and that is
data points is expressed as (k) and the average number of
shown in Tables 7 and 8.
detected failures is represented as N . N can be given as,
∑k
N = i=1 N(ti )∕k . ­R2 € [0, 1] measures the percentage of the
2.3.2 Relative error (RE)
total variation about the mean counted for the fitted curve.
The larger ­R2 is better the model explains the variation in
The capacity of SRGM model in predicting the behavior of
the data. When the approach reaches the value near to 1 it
failure from the failure observed is called predictive validity.
shows better effect (Jin and Jin 2016).
RE is selected as the predictive validity criteria. The RE is
formulated as,
M(t) − N(t) 3 Experiments and results
RE = . (16)
N(t)
3.1 Dataset description
In Eq. (16), where M(t) describes the estimated number
of failures and N(t) describes the observed number of fail- In the literature, there are seven datasets to test the soft-
ures. The relative difference between observed and estimated ware reliability growth model. The proposed chaotic grey
number of failures are estimated at the time (t).Positive value wolf optimization approach to compute the characteristics
of RE represents over-estimation while Negative values of of the software reliability growth model is tested on data-
RE represent under-estimation. The values closer to zero is set1 (DS1) and dataset2 (DS2) by the Tandem Computers
more accurate prediction. RE (t) = 0 fits better to the data Software Failure (Choudhary et al. 2017). In dataset1 (DS1),
prediction (Li et al. 2015). the software was tested for 18 weeks and during the testing
phase 51.63 CPU hours are consumed and 330 faults are
2.3.3 Coefficient of determination ­(R2) detected. Table 5 describes DS1. In dataset2 (DS2), the soft-
ware was tested for 18 weeks and during the testing phase
The capability is to check the fitting power of SRGM is a 10,000 CPU hours are consumed and 100 faults are detected.
correlation index of the regressive curve equation (R2). The Table 6 describes DS2.
ratio of the amount of square resulting from the tendency
model to that from the constant model is formulated: 3.2 Performance evaluation

The folk size of grey wolves is certainly taken as 30 and 110

iterations to ascertain the results based on the constricted

Table 1  Chaotic maps results Problem GW1 GW2 GW3 GW4 GW5 GW6
with constricted benchmark
functions on CGWO (GW1– Bernoulli map − 14.1857 − 0.52058 − 0.87665 − 32,350.4 52,672.1 − 6329.29
GW6)
Logistic map − 13.4120 − 0.667739 − 0.488866 − 32,236.7 39,974.8 − 6452.64
Chebyshev map − 14.7998 − 0.81434 − 0.9681 − 30,645.2 53,814.1 − 65,463.16
Circle map − 11.1283 − 0.21189 − 0.17956 − 31,212.6 186,242 − 6481.72
Cubic map − 14.4847 − 0.59969 − 0.54943 − 30,162.7 38,497.2 − 6634.46
ICMIC map − 11.1984 − 0.514268 − 0.105644 − 32,579.2 23,729.4 − 6589.12
Piecewise map 12.2842 − 0.620454 − 0.74163 − 31,275.1 56,104.5 − 6451.86
Singer map − 13.2749 − 0.40466 − 0.85799 − 30,202.2 64,423.1 − 6349.35
Sinusoidal map − 14.3466 − 0.33927 − 0.82258 − 30,022.5 43,857.1 − 6382.18
Tent map − 13.2748 − 0.43695 − 0.80427 − 30,494.2 26,421.2 − 6484.28
Sine map − 14.4705 − 0.51969 − 0.28356 − 32,195.2 26,868.2 6015.14
Gaussian map − 15.9102 − 0.567739 − 0.101344 − 33,250.4 55,231.2 − 6229.18

13
 P. Dhavakumar, N. P. Gopalan

Table 2  Chaotic maps results Problem GW7 GW8 GW9 GW10 GW11 GW12
with constrained benchmark
functions on CGWO (GW7– Bernoulli map 112.2438 − 0.05861 602.1739 6294.23 0.626 − 69.2839
GW12)
Logistic map 529.2649 − 0.076304 612.46 6045.14 0.635 − 68.1368
Chebyshev map 48.2278 − 0.09874 681.43 7046.13 0.672 − 47.3472
Circle map 52.3442 − 0.05812 676.67 027.24 0.650 − 76.3629
Cubic map 35.2463 − 0.07874 648.4 7038.43 0.639 − 143.363
ICMIC map 37.2641 − 0.09375 673.643 7063.12 0.668 − 103.368
Piecewise map 629.3459 − 0.09524 618.135 7054.24 0.652 − 213.368
Singer map 128.4740 − 0.05285 620.321 7046.26 0.625 − 408.478
Sinusoidal map 549.2327 − 0.03662 649.184 7013.17 0.638 − 2298.378
Tent map 828.4649 − 0.09466 663.12 7010.43 0.639 − 2538.36
Sine map 582.1472 − 503.093 612.37 7044.13 0.661 − 73.2839
Gaussian map 185.4121 − 0.06451 652.471 6842.24 0.648 − 576.4821

Fig. 2  Relative error (RE) curve

Fig. 3  Coefficient of determina-

tion ­(R2) curve

13
An efficient parameter optimization of software reliability growth model by using chaotic…

Fig. 4  Comparing five algo-

rithms with iterations

comparing with other four models can yield better predic-

Table 3  Number of failures at various intervals tion value.
Time Before After optimi- After optimi- After optimiza-
In Fig. 3, shows the curves of R ­ 2 versus time in weeks
optimiza- zation (PSO) zation (GWO) tion (CGWO) for each model. It is noticed that the Inflexion S-shaped
tion approach is the best fit approach to axis ­R2 is near to 1. With
the highest accuracy of prediction using increasing sample
1 15 14 11 6
size it is concluded that the proposed model by measuring
5 25 21 19 12
other four models Inflexion S-shaped approach can yield
10 46 35 29 21
better prediction coefficient. The fitness function comparison
15 18 17 15 8
with five algorithms is given Fig. 4.
20 33 26 22 12
In Table 3, the number of failures after the optimization
has shown vast difference and has good convergence fit-
ness. More than 70% of the failures are reduced in CGWO
algorithm optimization whereas GWO shows 40% and PSO
benchmarking functions, to scrutinize the quality of the shows 25% reduction performance.
envisaged CGWO heuristic. Compared with other optimi-
zation techniques including GWO, PSO, GA and WO, the
proposed algorithm accomplishes essential authenticity. For 4 Conclusion
GWO two random r1 and r2 vectors are in the range (0, 1),
with the control parameter being diminished progressively This paper envisages utilizing the CGWO heuristic as a
by 2 to 0. The range of (0, 1) and the governed variable is new approach to quantify SRGM attributes. Various bar-
reduced values from 2 to 0 in the course of incarnations are riers in the existing techniques are outsmarted by the pro-
taken spontaneously for CGWO and the chaotic function posed technique. Datasets have been applied to achieve
variables are a = 0, 5 and b = 0, 2. the results of evaluation criteria using parameter estima-
The outcome being that CGWO’s Chebyshev map shows tion technique. Compared to the GWO and PSO method,
the best outcomes in controlling benchmarking issues by proposed CGWO performs better integration and improves
outclassing all other maps. All the chaotic maps on the parameter tuning. In the Meta heuristic CGWO algorithm
constricted benchmarking functions are adhered for the the vital success is based on exploration and exploitation.
outcomes and the best map is determined. Out of 12 con- When optimization performed number of software fail-
straints 10 were performed best by Chebyshev map (shown ure decreases and increases the optimality. To enhance
in Tables 1 and 2). In Fig. 2, relative error and the time is the reliability of SRGM, the parameters of SRGM esti-
described in weeks for each model. The best fit approach is mated using TEF and enhanced using chaotic maps to
the Inflexion S-shaped approach to [RE = 0] axis. The pro- improve search performance. The result shows that in the
posed model by utilizing Inflexion S-shaped approach by proposed approach the prediction error decreases, uses

13
 P. Dhavakumar, N. P. Gopalan

characteristics of the data itself without assumptions, rea- comparatively get better results of parameter estimation of
sonable predicting capacity and all the prediction process SRGM and help in predicting software reliability.
is automated and does not require user participation. The
Presented chaotic maps have been helps to widely used
by the researchers, mathematicians in the optimization Appendix
field. It obviously helps to efficiently traverse the search.
In the future, we will continue to find some approach to See Tables 4, 5, 6, 7, 8.

Table 4  Chaotic maps Name of map Equation

{ }
Bernoulli map VP ∕0.6, 0 < VP ≤ 0.6
VP+1 =
(VP − 0.6)∕0.4, 0.6 < VP < 1
Logistic map VP+1 = 4VP (1 − VP )
Chebyshev map VP+1 = cos(0.5 cos −1VP )
Circle map VP+1 = VP + 0.5 − 1.1
sin(2𝜋VP ) mod (1)
𝜋
Cubic map VP+1 = 2.59VP (1 − VP2 )
Interact chaotic map with infinite collapses VP+1 = sin(70∕VP )
(ICMIC) map
{ }
Piecewise map VP ∕0.7, VP ∈ (0, 0.7)
VP+1 =
(1 − VP )(1 − 0.7), VP ∈ (0.7, 1)
Singer map VP+1 = 1.073(7.86VP − 23.31VP2 + 28.75VP3 − 13.302875VP4
Sinusoidal map VP+1 = 2.3VP2 sin(𝜋VP )
{ }
Tent map VP ∕0.4, 0 < VP ≤ 0.4
VP+1 =
(1 − VP )∕0.6, 0.4 < VP ≤ 1
Sine map VP+1 = sin(𝜋VP )
{ }
Gaussian map 0, VP = 0
VP+1 =
(1 − VP ) mod (1)VP ≠ 0

Table 5  Dataset1 (DS1) Table 6  Dataset2 (DS2)

Weeks Aggregate execution time Aggregate number Weeks Aggregate debugging time Aggregate number
(CPU hours) of detected faults (CPU hours) of detected faults

1 6.83 65 1 985 14
2 8.02 89 2 1981 28
3 11.65 106 3 2934 32
4 14.25 115 4 3742 48
5 17.89 140 5 4948 59
6 20.23 166 6 5416 67
7 21.47 188 7 5924 72
8 23.12 210 8 6691 77
9 26.92 238 9 7114 82
10 28.04 259 10 7519 86
11 30.49 272 11 7948 90
12 34.43 291 12 8059 93
13 38.51 302 13 8321 96
14 40.41 310 14 8543 98
15 42.47 322 15 8981 99
16 44.34 324 16 9327 100
17 47.58 327 17 9543 100
18 51.63 330 18 10,000 100

13
An efficient parameter optimization of software reliability growth model by using chaotic…

Table 7  Parameters comparison of TEFs by MSE for DS1

TEF Weibull Exponential Weibull Logistic Generalized logistic Inflexion S-shaped

MSE 0.56 0.71 0.93 1.04 0.23

Parameter N = 4.792, N = 786, b = 101.0, N = 78.54, N = 71.59, A = 204.0, N = 56.91, A = 156.3, α = 10
estimation b = 3002.0, m = 362.0, θ = 695.2 A = 04.13, α = 4988.0, k = 19.16
m = 115.1 α = 2926.0

Table 8  Parameters comparison of TEFs by MSE for DS2

TEF Weibull Exponential Weibull Logistic Generalized logistic Inflexion S-shaped

MSE 4.58 4.31 4.90 3.95 3.55

Parameter esti- N = 110.8; b = 0.0021; N = 122.7; b = 0.0139; N = 92.9; A = 25.01; N = 107; A = 1.003; N = 97.3;
mation m = 1.856 m = 1.471; θ = 1.242 α = 0.2274 α = 0.024 2; k = 5.127 A = 13.97;
α = 0.2601

References Kohli M, Arora S (2018) Chaotic grey wolf optimization algorithm for
constrained optimization problems. J Comput Des Eng 5(4):458–
472. https​://doi.org/10.1016/j.jcde.2017.02.005
Abd-Elkader AG, Saleh SM, Eiteba MM (2018) A passive islanding
Lakshmanan I, Ramasamy S (2015) An artificial neural-network
detection strategy for multi-distributed generations. Int J Electr
approach to software reliability growth modeling. Procedia Com-
Power Energy Syst 99:146–155. https​://doi.org/10.1016/j.ijepe​
put Sci 57:695–702. https​://doi.org/10.1016/j.procs​.2015.07.450
s.2018.01.005
Li Q, Li H, Lu M (2015) Incorporating S-shaped testing-effort func-
Cascone A, Manzo R, Piccoli B, Rarità L (2008) Optimization versus
tions into NHPP software reliability model with imperfect debug-
randomness for car traffic regulation. Phys Rev E 78(2):026113.
ging. J Syst Eng Electron 26(1):190–207. https:​ //doi.org/10.1109/
https​://doi.org/10.1103/PhysR​evE.78.02611​3
JSEE.2015.00024​
Cheng M, Wu G, Yuan M, Wan H (2016) Semi-supervised soft-
Li Z, Jing X, Zhu X (2018) Progress on approaches to software defect
ware defect prediction using task-driven dictionary learning.
prediction. IET Softw 12(3):161–175. https​://doi.org/10.1049/
Chin J Electron 25(6):1089–1096. https ​ : //doi.org/10.1049/
iet-sen.2017.0148
cje.2016.08.034
Mallikharjuna K, Anuradha K (2015) An efficient method for software
Choudhary A, Baghel A, Sangwan O (2017) Efficient parameter esti-
reliability growth model selection using modified particle swarm
mation of software reliability growth models using harmony
optimization technique. Int Rev Comput Softw 10(12):1169. https​
search. IET Softw 11(6):286–291. https ​ : //doi.org/10.1049/
://doi.org/10.15866​/ireco​s.v10i1​2.8089
iet-sen.2015.0171
Mirjalili S, Mirjalili S, Lewis A (2014) Grey wolf optimizer. Adv Eng
Cutolo A, De Nicola C, Manzo R, Rarità L (2012) Optimal paths on
Softw 69:46–61. https:​ //doi.org/10.1016/j.adveng​ soft.​ 2013.12.007
urban networks using travelling times prevision. Model Simul
Packianather MS, Yuce B, Mastrocinque E, Fruggiero F, Pham DT,
Eng. https​://doi.org/10.1155/2012/56416​8
Lambiase A (2014) Novel Genetic Bees Algorithm applied to
D’Apice C, Manzo R, Rarità L (2011) Splitting of traffic flows to con-
single machine scheduling problem. In: 2014 World Automation
trol congestion in special events. Int J Math Math Sci. https​://doi.
Congress (WAC). IEEE, pp 906–911. https​://doi.org/10.1109/
org/10.1155/2011/56317​1
wac.2014.69361​94
Decoderz (2019) Behavior of grey wolf optimization (GWO) algorithm
Rarità L, D’Apice C, Piccoli B, Helbing D (2010) Sensitivity analy-
using meta-heuristics method. Transpire Online. https:​ //transp​ ireo​
sis of permeability parameters for flows on Barcelona networks.
nline​.blog/2019/08/09/behav​ior-of-grey-wolf-optim​izati​on-gwo-
J Differ Equ 249(12):3110–3131. https​: //doi.org/10.1016/j.
algor​ithm-using​-meta-heuri​stics​-metho​d/. Accessed Sept 2010
jde.2010.09.006
Diwaker C, Tomar P, Poonia R, Singh V (2018) Prediction of software
Riaz S, Arshad A, Jiao L (2018) Rough noise-filtered easy ensemble
reliability using bio inspired soft computing techniques. J Med
for software fault prediction. IEEE Access 6:46886–46899. https​
Syst. https​://doi.org/10.1007/s1091​6-018-0952-3
://doi.org/10.1049/iet-sen.2014.0108
Fera M, Fruggiero F, Lambiase A, Macchiaroli R, Todisco V (2018) A
Roy P, Mahapatra G, Dey K (2017) An efficient particle swarm opti-
modified genetic algorithm for time and cost optimization of an
mization-based neural network approach for software reliability
additive manufacturing single-machine scheduling. Int J Ind Eng
assessment. Int J Reliab Qual Saf Eng 24(04):1750019. https​://
Comput 9(4):423–438. https​://doi.org/10.5267/j.ijiec​.2018.1.001
doi.org/10.1109/ACCES​S.2018.28653​83
Jin C, Jin S (2016) Parameter optimization of software reliabil-
Singh L, Tripathi A, Vinod G (2015) Approach for parameter esti-
ity growth model with S-shaped testing-effort function using
mation in Markov model of software reliability for early predic-
improved swarm intelligent optimization. Appl Soft Comput
tion: a case study. IET Softw 9(3):65–75. https​://doi.org/10.1109/
40:283–291. https​://doi.org/10.1016/j.asoc.2015.11.041
TNSM.2018.28481​05
Kim T, Lee K, Baik J (2015) An effective approach to estimating the
Vizarreta P, Trivedi K, Helvik B, Heegaard P, Blenk A, Kellerer W,
parameters of software reliability growth models using a real-
Mas Machuca C (2018) Assessing the maturity of SDN controllers
valued genetic algorithm. J Syst Softw 102:134–144. https​://doi.
with software reliability growth models. IEEE Trans Netw Serv
org/10.1016/j.jss.2015.01.001

13

Manag 15(3):1090–1104. https:​ //doi.org/10.1109/TR.2018.28049​
22
Wu F, Jing X, Sun Y, Sun J, Huang L, Cui F, Sun Y (2018) Cross-
project and within-project semisupervised software defect predic-
tion: a unified approach. IEEE Trans Reliab 67(2):581–597. https​
://doi.org/10.1109/TR.2018.28049​22
Yang X, Tang K, Yao X (2015) A learning-to-rank approach to soft-
ware defect prediction. IEEE Trans Reliab 64(1):234–246. https​
://doi.org/10.1016/j.asoc.2016.08.006
Yazdanbakhsh O, Dick S, Reay I, Mace E (2016) On deterministic
chaos in software reliability growth models. Appl Soft Comput
49:1256–1269. https​://doi.org/10.1007/s1229​3-017-0247-0
Yu Y, Gao S, Cheng S, Wang Y, Song S, Yuan F (2017) CBSO: a
memetic brain storm optimization with chaotic local search.
Memet Comput 10(4):353–367. https ​ : //doi.org/10.1109/
TR.2016.25705​57
13
P. Dhavakumar, N. P. Gopalan
Zeephongsekul P, Jayasinghe C, Fiondella L, Nagaraju V (2016)
Maximum-likelihood estimation of parameters of NHPP soft-
ware reliability models using expectation conditional maximiza-
tion algorithm. IEEE Trans Reliab 65(3):1571–1583. https​://doi.
org/10.1109/TR.2016.25705​57
Zhu M, Pham H (2018) A two-phase software reliability mod-
eling involving with software fault dependency and imperfect
fault removal. Comput Lang Syst Struct 53:27–42. https​://doi.
org/10.1016/j.cl.2017.12.002
Publisher’s Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations.