Test case prioritization using firefly algorithm for software testing

Received August 14, 2019, accepted September 3, 2019, date of publication September 10, 2019,
date of current version September 25, 2019.
Digital Object Identifier 10.1109/ACCESS.2019.2940620
Test Case Prioritization Using Firefly
Algorithm for Software Testing
MUHAMMAD KHATIBSYARBINI 1, MOHD ADHAM ISA1, DAYANG N. A. JAWAWI1,
HAZA NUZLY ABDULL HAMED 1, AND MUHAMMAD DHIAUDDIN MOHAMED SUFFIAN 2
1School of Computing, Faculty of Engineering, Universiti Teknologi Malaysia, Johor Bahru 81310, Malaysia
2Business Solution and Services, MIMOS Technology Solutions Sdn. Bhd., Kuala Lumpur 57000, Malaysia
Corresponding author: Muhammad Khatibsyarbini (fkmuhammad4@gmail.com)
This work was supported in part by the Fundamental Research Grant Scheme, vote number 5F069, under the Ministry of Education
Malaysia, and in part by the Research University Grant Scheme, vote number 01M55, under Universiti Teknologi Malaysia.
ABSTRACT Software testing is a vital and complex part of the software development life cycle. Optimization
of software testing is still a major challenge, as prioritization of test cases remains unsatisfactory in terms
of Average Percentage of Faults Detected (APFD) and time execution performance. This is attributed
to a large search space to find an optimal ordering of test cases. In this paper, we have proposed an
approach to prioritize test cases optimally using Firefly Algorithm. To optimize the ordering of test cases,
we applied Firefly Algorithm with fitness function defined using a similarity distance model. Experiments
were carried on three benchmark programs with test suites extracted from Software-artifact Infrastructure
Repository (SIR). Our Test Case Prioritization (TCP) technique using Firefly Algorithm with similarity
distance model demonstrated better if not equal in terms of APFD and time execution performance compared
to existing works. Overall APFD results indicate that Firefly Algorithm is a promising competitor in TCP
applications.
INDEX TERMS Firefly Algorithm, metaheuristic, software engineering, artificial intelligence, search-based
software testing, test case prioritization.
I. INTRODUCTION
In the software development process, software testing has a
long execution time and can be the most expensive phase [1].
Software testing is arguably the least understood part of
the software development process. Even software testing is
executed repeatedly, it is often carried out in haste, due to
time constraints and fixed resources. In light of this, it has
been reported that Test Case Prioritization (TCP) application
appears to enhance test viability in software testing activ-
ity [2]–[5]. Test case prioritization approach was first men-
tioned by [6]. That work however only applied prioritization
on test cases that had gone through test case selection. Two
authors, Rothermel and Harold proposed and evaluated the
approach in a further broad context. Considering a test suite
as described in Table 1 [7]. Note that this study example
depicts an ideal situation in which fault detection information
is known.
The associate editor coordinating the review of this manuscript and
approving it for publication was Muhammad Afzal.
TABLE 1. Test suite example.
TCP aims to order a set of test cases to achieve early
optimization based on preferred properties [2], [8]. The goal
of prioritization is to maximize early fault detection. It is clear
that by ordering the test case as TC5-TC3 is far superior than
any other combination, as TC5-TC3 order detects all of the
faults at an earlier rate as can be seen on Table 1. In real
world problems, it is often difficult to determine which tests
will actually reveal faults. Hence, the test case prioritization
steps depend on several methods, expecting that an early
intensification of a certain chosen method will end in boost-
ing of earlier fault discovery. There are many dimensions of
the test case prioritization approach. As many as eight broad
132360
This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see http://creativecommons.org/licenses/by/4.0/
VOLUME 7, 2019

M. Khatibsyarbini et al.: Test Case Prioritization Using Firefly Algorithm for Software Testing
dimensions h described by [9]. Those TCP techniques were
based on their commonalities in selection procedures, input
type, and output type.
TCP can be achieved by utilizing string metrics which are
able to distinguish test cases in terms of character differences,
followed by prioritization meeting a predefined fitness func-
tion. In recent years, existing works appear to have prioritized
test cases using only information related to generated test
cases [5], [10]–[12]. For instance, by relying on available
information such as test case input, software testers may
prioritize test cases prior to the availability of system source
code. This reduces overall testing time. TCP can be initi-
ated right after the design phase. Among the numerous TCP
techniques proposed, there remains room for improvement in
terms of time execution performance [4], [5]. This problem
is not only widespread on non-artificial intelligence appli-
cations in TCP, but also on most Artificial Intelligence (AI)
applications in TCP, necessitating improvements especially
in terms of APFD and execution time, as indicated by recent
studies [3]–[5], [13], [14].
Most AI applications in TCP begin with identifying and
differentiating test cases first before prioritization is carried
out [10], [15], [16]. A strategy which can be used to dif-
ferentiate test cases is calculating character differences via
string metrics. Once character differences among test cases
are established, prioritization can be pursued. String metrics
play an important role in textual similarity research. Applica-
tions in tasks such as information retrieval, text classification,
document clustering, topic detection, topic tracking, ques-
tions generation, question answering, essay scoring, short
answer scoring, machine translation, and text summarization
all utilize string metrics [17]. String metrics can be fur-
ther categorized based on their metric calculation strategies.
In calculating the distance between test cases, specificity and
reliability of string distances in the prevailing prioritization
goal are needed. In existing works in [3], [5], [10] only one
string distance was utilized, which may not resolve redun-
dancy issue where test cases have widespread equidistant
weights. This problem may also lead to lowered measures of
fault detection at the end of the TCP process.
Upon completion of character difference calculation
among test cases using string metrics, an effective prioriti-
zation algorithm is then needed to arrange test cases based
on respective character difference weights among test cases.
Recent works by [3], [18]–[20] indicate that a Swarm Intel-
ligence (SI) algorithm can have a significant effect on TCP
process. Additionally, as indicated by findings in [4], the use
of artificial intelligence algorithms may yield better APFD
results as well as improved execution time performances.
However, the uses of SI algorithm such as Firefly Algo-
rithm was not being explored in TCP context which can
benefit with the utilization of string metrics.
From the recent studies we can see that there were sev-
eral problems that arise which worth exploring. To have an
increment of APFD result as well as the time execution time,
an AI technique is suggested and also recent study shows
SI algorithm could have a significant result. Consequently,
macro research question or problem of this paper is, ‘‘Which
SI algorithm would perform better in TCP context’’. Since,
Firefly Algorithm was not being explored in TCP, authors has
decided to implement the algorithm.
Work in [19] explains the implementation of Firefly Algo-
rithm in generating test cases. The work manages to outper-
form Ant Colony Optimization (ACO) algorithm. The goal
of test case generation is distinguishable from TCP whereby,
in test case generation, prioritization of test cases is initiated
first, followed by selection of only required test cases. Mean-
while, TCP only involves prioritization of test cases, focusing
on ordering of test cases, according to a predefined goal.
The work suggests the application of Firefly Algorithm on
relevant software testing procedures by means of optimizing
the best possible test sequences. Motivated by the suggestion,
the schematic of firefly schematic algorithm implemented
in the work is possible to be extracted and revised for appli-
cation in TCP context. Therefore, the research objectives of
this paper were.
• To explore prioritizing an optimal ordering of test cases
by using Firefly Algorithm, which has not been reported
in existing TCP literature.
• To apply the Firefly Algorithm to four different bench-
mark programs aided with string metrics.
• To compare performance in terms of APFD and execu-
tion time performance against existing works.
The rest of the paper is structured as follows.
Section 2 describes related works in TCP using artificial
intelligence algorithms with string distances. In Section 3,
an overview of Firefly Algorithm is outlined. In Section 4,
the application of Firefly Algorithm in TCP is presented
and illustrated. In Section 5, an empirical evaluation of the
proposed work in comparison with existing works is, per-
formed with results and discussion included. Lastly, Section 6
provides a conclusion.
II. RELATED WORK
Regression testing is performed to confirm that alterations
have not impacted previously functioning software [21], [22].
As software evolves, test suites have a tendency to grow in
size, eventually requiring costly resources to execute. Despite
the costs, software testing remains a crucial stage to be exe-
cuted in order to ensure products built are exactly as intended.
Therefore, the principle objective of performing regression
test is to guarantee that any alterations acted upon a software
system will not influence the unaltered components. When
a program has been modified, it is necessary to ensure that
any altered parts do not adversely affect unaltered parts of
the software, such as the functionalities of that particular
software version. In such context, test cases constructed for
that software version may only yield accurate outputs for that
particular version alone. Those test cases may yield incor-
rect outputs for subsequent versions of the software created
through alterations performed on initial software version.
Here, regression testing may prove advantageous.
VOLUME 7, 2019 132361

FIGURE 1. Schematic view of firefly algorithm.
During a regression test, changed parts in the software
are checked using a regression test technique, in order to
check if the unchanged parts of the software behave func-
tionally similar with the original version of the software.
Past research indicated that regression testing is an expensive
process which may consume more than 33% of the total
cost of the software [23]. In [24], various approaches are
examined to augment the importance of accumulated test
suites in regression testing. A large and complex software
system will typically involve a huge pool of test cases to cover
all functionalities.
Within regression testing, TCP operates by prioritizing test
cases through certain means utilizing only available infor-
mation. Work in [5] tried to maximize diversity among test
cases utilizing test case input information. This is in contrast
with work in [10], in which each test case is assigned a
distance weight in connection to other test cases, utilizing
a string metric such as edit distance, which computes char-
acter difference among test cases. These weighted test cases
are subsequently prioritized using a predefined algorithm.
Regardless of the strategies used, the main goal is fixed to
give higher priorities to test cases that are largely dissimilar
(i.e., invoke different methods and possess large numerical
values of character difference), consequently maximizes test
case diversity, which in turn casts a high possibility in unique
fault detection [25].
The implementation of artificial intelligence in TCP is
not limited to any specific strategy [4]. Within TCP itself,
there several algorithms have been used including Genetic
Algorithm (GA) [26]–[29], [31]–[34], Greedy, [35], [36],
Particle Swarm Optimization [37], [18], and others [5], [13],
[38]–[40]. Recent work by Jiang and Chan [5] has indicated
that a heuristic prioritization algorithm can have a significant
effect on the TCP process. In their finding, the use of artificial
intelligence algorithm may increase the yield of average
percentage of faults detected (APFD) measure. However,
the work did not compare their proposed prioritization algo-
rithm against existing works that have utilized AI algorithms.
A comparison study between the proposed prioritization
algorithm and existing AI algorithms in TCP would be bene-
ficial to researchers and real world applications, as such study
would produce comparative performances in terms of APFD
and execution time. Motivated by this, any efforts directed
toward fulfilling such beneficial insights which in turn may
enhance TCP capabilities are worth exploring.
In [41], nearest neighbor algorithm was used. The approach
aimed to reduce the distance between neighboring test cases
in the execution order. Distance was quantified as the number
of different instructions executed between two test cases. The
results from the experiment showed that the order of test
case execution instructions is crucial toward yielding a better
APFD result. However, the work reported that prioritization
execution time utilizing nearest neighbor algorithm yielded
the worst performance time. Further, the authors suggest
an exploration of possible existing alternative algorithms in
TCP, which might yield better performance with respect to
program and test suite size.
The Swarm Intelligence (SI) algorithm application in [18]
has been reported to offer an overall performance improve-
ment using PSO in a multi-objective TCP. Further, the work
reports that PSO prioritization was able to minimize total
test cases required to achieve 100% coverage at a shorter
execution time. Meanwhile, in [37], PSO implementation
in TCP indicated an improvement in APFD result as com-
pared to random ordering. Additionally, the study reports
the ability of PSO in TCP process toward improving fault
detection capability in the early stages of testing. However,
the approach provided limited discussions concerning PSO
performance in comparison to other SI algorithms in TCP.
III. OVERVIEW OF FIREFLY ALGORITHM
Light flashes from a firefly draw nearby fireflies. The light
flashes can be formulated through associating them with an
objective function to be optimized, which in turn makes it
possible to formulate a new optimization algorithm. Figure 1
shows a schematic view of the Firefly Algorithm. There are
132362 VOLUME 7, 2019

three assumptions in the Firefly Algorithm discussed in [19]
as follows:
a. All fireflies are unisexual. Every firefly either attracts
or gets attracted to each of other fireflies.
• All fireflies will get attracted to any fireflies with-
out any discrimination.
• For example, there are five fireflies and each of
them will get attracted to every single firefly avail-
able as they approach one another.
b. The attractiveness of a firefly is directly proportional to
the brightness of the firefly.
• As fireflies are attracted to others available,
the brightness of the firefly becomes a priority
among them to rank attractiveness.
c. Fireflies move randomly if they do not find a more
attractive firefly in adjacent regions.
• If there are two or more fireflies with same bright-
ness, firefly will random randomly move toward
either one.
FA has been applied in spatial fields consisting of different
dimensions with promising efficiency and superiority over
other algorithms [19], [20]. FA is a metaheuristic algorithm,
which assumes that a solution of an optimization problem
is encoded as the location of an agent/firefly, while the
objective function is encoded as light intensity. Within the
Firefly Algorithm are two crucial considerations: the variant
of light intensity, sometimes stated as brightness, and formu-
lation of the interaction among fireflies. For simplicity, it is
assumed that the spectacle of a firefly is decided by way of its
brightness, which in turn is associated with an encoded goal
characteristic. Simply put, the attractiveness of a firefly in a
search space is directly proportional to the objective function
value of the firefly.
The schematic view of the Firefly Algorithm to be applied
in TCP, with considerations of assumptions described earlier,
is shown in Figure 1. First, the algorithm starts with an objec-
tive function derivation. Next the calculation of the adjacency
of distance matrix between firefly agent (FA) and its bright-
ness which is encoded to determine the attractiveness of each
firefly. Subsequent movement of a firefly will be based on
brightness value. The movement stops once all fireflies have
been visited. All movements are recorded. Finally, the best
sequence of fireflies is chosen based on shortest distance.
IV. APPLYING FIREFLY IN TCP
In the previous section, a brief overview of the basic Firefly
Algorithm concept, motivation of Firefly Algorithm in soft-
ware testing, and a schematic view of the algorithm in TCP
context were provided. For clarity, in this section, the flow
of Firefly Algorithm application will be described in detail.
Table 2 show a firefly itself represents a test case in TCP. The
objective function that defines a firefly attractiveness repre-
sents a test case similarity weight and uniqueness. All move-
ments of the fireflies are subsequently recorded. The best path
is consequently designated the best test suite sequence.
TABLE 2. Firefly algorithm component mapping.
FIGURE 2. Schematic view of test case distance adjacency.
Based on Table 2, the representation of firefly components
is illustrated in Figure 2 below. Giving there are five test cases
to be prioritized, the probability of having the best prioritized
arrangement is one over five factorials (1/5!). From Figure 2,
for each test case will act as firefly agent while the distance
between each test cases denotes the attractiveness function
among firefly agent. To find the best-prioritized arrangement,
the Firefly Algorithm is used in this experiment with the
string metric as a fitness function.
In this paper, the actual result of distance calculated for
each benchmark program will not be illustrated, as the num-
ber of test cases and their respective contents are too large
and too long to be presented adequately. Instead, to demon-
strate how distance is calculated in this research work, five
dummy test cases were created. Table 3 shows five dummy
test cases created to demonstrate distance calculation and
its weight using ‘‘Term Frequency–Inverse Document Fre-
quency’’ (TD-IDF). For the distance among test cases, Edit
distance metric is used.
Edit distance involves the calculation of string charac-
ters, where, the quantified value represents the least number
of insertions, deletions, and replacements to transform first
string into second string. Edit distance is fixed to involve only
a minimal transformation of first string into second string.
For example, the edit distance between "9555" and "85577"
VOLUME 7, 2019 132363

TABLE 3. Five dummy test cases.
TABLE 4. Adjacency matrix for distance between test cases.
is 3, since three edits are needed to transform ‘‘9555’’
into ‘‘85577’’. The following edits illustrate edit distance
calculation:
1) 9555 → 8555 (substitution of ‘‘8’’ for ‘‘9’’);
2) 8555 → 8557 (substitution of ‘‘7’’ for ‘‘5’’);
3) 8557 → 85577 (insertion of ‘‘7’’ at the end).
In TFIDF, term frequency, tf, calculation starts first with
the use of selected term or string frequency in a document.
Simply put, tf refers to the amount of term t occurring within
a document d. Meanwhile, the formula for inverse document
frequency, idf, deals with the significance of a term in a pool
of documents. Subsequently, TFIDF equation is expressed as
follows.
TFIDF(t) = t/T × logN
/nt
where: t = term frequency in one document; T = term
frequency in all documents; N = number of term occurrence
throughout document; and nt = number of documents has
term t. These two string metrics were then used to calcu-
late the distance and weight of the dummy test cases in
Table 3 below.
The results of calculated distance for dummy test cases
in Table 3 are shown in Table 4. All string distances were
quantified based on percentage of similarity from scale 0 to 1.
Value 1 denotes the highest differences of two test cases
while, value 0 denotes zero distance of two test cases. For
the test case weight which will represent as the bright-
ness of each test cases, the value of all test cases weight
will be similar among each other using TD-IDF formula-
tion as all have different single string. TF-IDF is a stand-
out scheme amongst most typically utilized term weighing
schemes in data retrieval methods. Attributed to this capa-
bility, TF-IDF has been regularly applied in experimental
investigations [42]. The weight among test cases was used to
assign brightness among firefly agents in Firefly Algorithm.
The value would be calculated as the product of one multiple
by one with log five over one which will have the value
TABLE 5. Test cases movement and update.
of 0.6989. Therefore, to make it easier to demonstrate the cal-
culation within the Firefly Algorithm, we assume the weight
of all test cases which act as the brightness of firefly agent
as 1. Table 5 show the test cases initial point and distance
travelled in every movement.
From Table 5, the selection of the next test cases in
movement update were based on the weight of the test case
which act as the brightness over the distance between the test
cases, where the highest will be selected as the next move.
In mathematical formulation, the brightness of a test cases
over distance calculation can be denoting as:
Weight of current Test Case
Distance to Next Test Case
Since the weight of all dummy test cases were all the same,
shorter distance were the only metric considered. Based on
Table 5, four starting points of movement end up with TC3 as
the last point to visited, and the lowest distanced travelled is
1.80 which tied with three best different sequences. This is
due to the similar weight of all dummy test cases. However,
in a real case study or benchmark program, the similar weight
problem diminishes and distances would be more dynamic.
To summarize the implementation of Firefly Algorithm in test
case prioritization, Figure 3 illustrates the flowchart of the
Firefly Algorithm used.
Based on Figure 3, the data were parsed from benchmark
programs and test case were extracted from the dataset. Then
calculation of test cases distances was performed by using
edit distance and with using TFIDF string metric. The test
case adjacency was then generating a matrix table as similar
to the dummy adjacency matrix in Table 4. Prioritization then
started with the first movement of an FA toward the brightest
non-visited test case until it reached the last firefly in the
search space. The overall path travelled would represent a
prioritized test case ordering. The shortest distance of full
sequence of the test cases is considered as the best path.
V. EMPIRICAL SETUP
In experiment setup phase, the experiment context, variables,
datasets, and design to validity are defined as follows.
132364 VOLUME 7, 2019

FIGURE 3. Flowchart of applied firefly algorithm.
A. EXPERIMENT CONTEXT
The experiment was performed in a controlled laboratory
setting. Two machines were used to run their experiment,
the first one is a Linux-based machine, run on a computer with
4GB RAM and Core i5 Intel processor. This first machine
was used to extract and execute test cases from each of
the three benchmark Siemens TCAS benchmark dataset on
a Linux environment. The second machine, equipped with
Intel i7 and 16 GB RAM, is used to run permutation of
test cases using four string metrics, implemented in Java
NetBeans 8.0.2 and measure experiment prioritization timing
performance, which required higher processing power as the
test case dimensionality matrix is large, with the test suite size
standing at 1052 × 1052 cells.
B. EXPERIMENT VARIABLES
The experiment variables comprise of dependent variable and
controlled variable. The dependent variable is an observed
element which is affected by changes made to the controlled
variables. Controlled variable refers to elements that the
authors alter to induce behaviors that are observable. The two
sets of variables are as follows:
Controlled variables:
• String metrics: Edit Distance and TFIDF
• Siemens TCAS Benchmark Program
• Real-world UNIX Benchmarks Programs
TABLE 6. Benchmark suite.
Dependent Variables:
• Average Percentage of Faults Detected (APFD)
• Time Execution
For benchmark applications, three real-world UNIX
benchmarks programs and three SIEMENs benchmarks
programs were used as test suites (downloaded from
http://sir.unl.edu) [43]. These benchmark programs are
popular among researchers [5] in search-based software
testing, especially in TCP. In order to assess the overall
performance and consistency of the proposed applied algo-
rithm in TCP process, we ran the prioritization process
of each algorithm with 30 iterations for every benchmark
program. The 30 iterations were intentionally being chosen
to the consistency of each algorithm since every single iter-
ation were made might resulted in slightly different results.
Table 6 shows the details of benchmark applications chosen
for this study.
From Table 6, the dataset used is TCAS which is originally
a C program of an aircraft collision avoidance system. It takes
12 integer inputs and produces one output. The program
comes with one base version and 41 faulty versions with
1608 test cases. The fault matrix is produced by executing all
test cases on all 41 faulty versions and compared against their
base version. Simple example of fault matrix can be referred
from Table 1. For TCAS program, the test cases do not lie
in any folder. The test case was written in a test script itself.
Figure 4 shows the main function for TCAS C program, while
Figure 5 shows the extracted test case from the test script.
As can be seen in figure 4, there were 12 variables declared
which gain its value from retrieved argument from the inputs.
Therefore, in Figure 5, there were also 12 columns where
each column contains the value input for each variable in
TCAS main function. The TCAS program were then used
for the experiment process design. Each of our benchmark
program accepts a file as an input and a string of inputs
entered from a command line terminal. Input data from
each file along with its respective command line content
was mined, converted it into strings, and calculated for its
similarity distance and uniqueness by using an edit dis-
tance. Based on the distance calculated, the test case finally
will be prioritized. Table 7 lists existing algorithms which
VOLUME 7, 2019 132365

FIGURE 4. Main function of TCAS C Program.
FIGURE 5. Example inputs for each test cases for TCAS C program.
TABLE 7. Prioritization techniques.
were used to compare their prioritization performances
against our proposed Firefly Algorithm.
From Table 7, Particle Swarm Optimization (PSO) is cho-
sen as one of comparable algorithm because, it is a quite sim-
ilar swarm intelligence algorithm. Apart from that, PSO also
has been widely used in TCP area. Local Beam Search (LBS)
on the other hand is one of latest algorithm implemented
in TCP and has quite convincing result. While for Genetic
Algorithm (GA) and Greedy, both of them were chosen as
they were the common technique used in TCP apart from ran-
dom. Table 8, summarize the reasoning behind the selection
of selected algorithm for this experiment.
C. EXPERIMENT FLOW
Figure 6 illustrates an overview of TCP experiment flow.
The flow is divided into four phases; (1) Information Extrac-
tion phase, (2) String Distance and TFIDF calculation for
TABLE 8. Selected algorithm comparative analysis.
FIGURE 6. Overview of experiment flow.
fireflies’ distances and brightness, (3) Prioritization phase,
and (4) Evaluation phase.
In phase 1, test cases were extracted from repository with
their input contents collected. Extracted input content for
each test case was placed on a separate text document to
ease calculation process in the next phase. Then, the orig-
inal program and versioning programs were compiled and
executed. Outputs produced from the original program were
compared against versioning programs. Test cases which
revealed faults, indicated by test cases which yielded different
outputs when ran on the original program and when ran on
versioning programs, were recorded. The record was kept in
the form of a fault matrix sheet.
In phase 2, character difference between test cases was
calculated based on extracted input contents. Computed
weights were subsequently populated into a test case distance
matrix. The matrix placed similarity and dissimilarity weights
and uniqueness of test cases in their respective positions.
132366 VOLUME 7, 2019

TABLE 9. Fault matrix for distance between test cases.
The weights and uniqueness are represented the firefly attrac-
tiveness values in the proposed Firefly Algorithm.
In phase 3, the proposed Firefly Algorithm was executed to
prioritize test suites, where the shortest possible cumulative
test case distance was selected as the best test case ordering,
which reflects the best efficiency in terms of execution time.
A complete flow of the proposed Firefly Algorithm applica-
tion in TCP is illustrated in Figure 3. The flowchart covers
extraction phase until prioritization phase. The best path trav-
elled by fireflies represents the best sequence of prioritized
test cases. Finally, in phase 4, prioritized test suites were
evaluated by calculating their APFD measures based on fault
matrix of their respective programs, obtained in phase 1. Time
execution for each prioritization technique was also recorded
to evaluate time execution efficiency.
D. EVALUATION METRIC
Average percentage fault detection rate (APFD) metric mea-
surement was used to quantify the rate of fault detection [8].
APFD has been widely used in numerous TCP experiments.
APFD is a metric used to quantify how rapid an optimized
ordering of test cases can discover faults [2], [7]. The results
of APFD values in this study ranged between zero to 100,
where a greater value indicates a better fault revealing rate.
The equation for calculating APFD value is shown as follows:
APFD = 1 −
TF1 + TF2 + . . . + TFn
n × m
+
1
2n
where T is a test suite containing n test cases and F is a set
of m faults revealed by T. TF1 is the first test case in T0
ordering of T which reveals fault i. The APFD value of T0
is calculated by using the APFD equation. To demonstrate
the calculation of APFD, one of the best sequences from
dummy Test Cases in Table 5 is used. The sequence selected is
TC2 – TC5 – TC1 – TC4 – TC3. Giving that the fault matrix
from extracted output as in Table 9 below.
Based on the Table 9 and APFD equation, the mathematical
calculation for the APFD would be as follow:
APFD = 1 −
1 + 1 + 1 + 2 + 2
5 × 5
+
1
2(5)
Therefore, the value of APFD for the prioritized sequence
for dummy dataset is 0.82.
VI. RESULT AND DISCUSSION
The experiments began with calculation of string distances
using Edit distance for all three benchmark applications.
FIGURE 7. APFD for FLEX program.
TABLE 10. Overall APFD assessments for GZIP.
Upon completion of string distance calculation, Firefly Algo-
rithm was applied. Prioritization performance of the proposed
Firefly Algorithm was subsequently compared with PSO,
GA, Local Beam Search (LBS) and Greedy.
A. THE APFD RESULTS
In this section, APFD results for each benchmark program is
illustrated using boxplots as shown in Figure 7. Overall APFD
assessment for each benchmark program is also tabulated as
presented in Table 9. All five benchmark programs are char-
acteristically distinguishable in terms of input data despite
stark differences in terms of program sizes. The discussion
of Firefly Algorithm in comparison to other algorithms are
provided at the end of this section.
Figure 7 shows a boxplot for APFD results while
Table 10 shows overall APFD assessment for FLEX bench-
mark program. Cross-referring the table and boxplot, it can
be concluded that Firefly Algorithm shows a better APFD
result in FLEX test application. The result shows Firefly
Algorithm is slightly better compared to other compared
algorithms, it still shows superior mean, median, and standard
deviation values. However, for max values, LBS somehow
manage to get the highest values with 0.9629. Even so, all
other values already indicate that the Firefly Algorithm is able
to produce consistently higher APFD signifying its applica-
bility in FLEX benchmark program context.
Figure 8 shows a boxplot for APFD results while
Table 11 shows the overall APFD assessment for GREP
benchmark program. Quite similar to the previous dataset,
VOLUME 7, 2019 132367

FIGURE 8. APFD for GREP program.
TABLE 11. Overall APFD assessments for GREP.
FIGURE 9. APFD for GZIP program.
TABLE 12. Overall APFD assessments for GZIP.
by cross-referring the table and boxplot, it can be concluded
that Firefly offers a better APFD result in GREP application.
Figure 9 shows a boxplot for APFD results, while
Table 12 shows the overall APFD assessment for GZIP
benchmark program. Again, similar to the previous dataset,
by cross-referring the table and boxplot, it can be con-
cluded that Firefly offers a better APFD result in GZIP
application. Overall Firefly performed better in all UNIX
FIGURE 10. APFD for TCAS program.
TABLE 13. Overall APFD assessments for TCAS.
TABLE 14. Overall APFD assessments for CS-TCAS.
Benchmarks Programs over 30 iterations. As for Siemens
program, Figure 10 and Table 13 articulate the results
obtained.
From Figure 10 and Table 13, once again a similar con-
vincing result from Firefly Algorithm is obtained. However,
in TCAS program, it is quite interesting to see that almost all
algorithms tend to have a nearly consistent value for each
iteration. Even so, by cross-referring the table and boxplot,
it can be concluded that Firefly shows a better APFD result
in TCAS application.
From Figure 11 and Table 14, once again a similar
convincing result from Firefly Algorithm is obtained. How-
ever, CS-TCAS starts to show inconsistent results while GA
has more consistency based on the std-dev value. Even so,
by cross-referring the table and boxplot, it can be con-
cluded that Firefly shows a better APFD result in CS-TCAS
application.
As for final J-TCAS, Figure 12 and Table 15, once again
show a quite similar result for the Firefly Algorithm over-
coming the other algorithms. However, similar to CS-TCAS,
J-TCAS which is the largest program, the std-dev values
show an increment also since both of them are large dataset.
132368 VOLUME 7, 2019

FIGURE 11. APFD for CS-TCAS program.
TABLE 15. Overall APFD assessments for J-TCAS.
FIGURE 12. APFD for J-TCAS program.
Even so, it is still acceptable, and Firefly shows a better APFD
result in J-TCAS application. In order to see the consistency of
APFD values, Figures 13 -18 show the graphs of all iteration
values of APFD for each dataset.
From Figure 13-18, Firefly Algorithm indicates a higher
consistency and suitability to be applied in TCP domain as
APFDs value obtained at each iteration, across three pro-
grams, did not show any sharp fluctuation. Any sharp fluc-
tuation is unfavorable as this may indicate a ‘‘by chance
occurrence’’. Firefly strategy has the best overall APFD result
as the selection of subsequent firefly path is solely driven
by next brightest firefly agent. In cases of more than one
agent with equal brightness, firefly path will randomly select
a firefly agent. LBS on the other hand looks promising but the
algorithm hardly depends on a constant width value where
different constant value set will have a significant impact
FIGURE 13. APFD graph for FLEX benchmark program.
FIGURE 14. APFD graph for GZIP benchmark program.
FIGURE 15. APFD Graph for GREP benchmark program.
on APFD results. Meanwhile, GA exhibits a unique behav-
ior attributed to its adaptive search, which may vary prior-
itization APFD score, as the performance highly depends
on its mutation fitness function. For Greedy algorithm, the
VOLUME 7, 2019 132369

FIGURE 16. APFD graph for TCAS benchmark program.
FIGURE 17. APFD graph for CSTCAS benchmark program.
FIGURE 18. APFD graph for JTCAS benchmark program.
algorithm searches for next agent which covers the furthest
distance which sometimes produces a sub-optimal solution
path. Overall, firefly strategy has the best overall APFD result
as the selection of the next path is solely driven by next
TABLE 16. Average execution time for benchmark programs.
brightest agent. The strategy also optimizes ordering of test
cases, by promoting test case diversity, as test cases with least
similarity are chosen first, which in turn may provide a higher
probability in detecting faults earlier.
B. THE EXECUTION TIME RESULTS
In terms of algorithm efficiency, Table 16 show the aver-
age time taken for all prioritization algorithms to complete
searching for solution paths in all five benchmark programs.
Firefly Algorithm, if not equal, outperforms LBS in terms
of average time taken to prioritize test cases in flex bench-
mark program, while greedy comes as second best. In gzip
benchmark, LBS strategy outperforms other algorithms while
firefly comes as the second fastest prioritizing algorithm.
In grep, firefly outperforms all prioritization algorithms, with
execution time 220 seconds, outperforming LBS by eight
seconds. For J-tcas the biggest benchmark program which the
biggest program among all four program used, Firefly have
the fastest time execution with 650, 100 seconds faster than
LBS. This result may be attributed to firefly strategy, which
solely focuses on next brightest firefly agent in searching,
resulting in a reduced search time. In addition, the number
of iterations were also fixed based on the sizes of benchmark
programs.
C. THE OVERALL RESULTS AND DISCUSSION
Table 17 summarizes overall results of prioritizing algorithms
across three benchmark programs. From Table 13, for grep
benchmark program, firefly outperforms other search strate-
gies, yielding the highest mean of APFD value (0.9517). LBS
comes as second best strategy while PSO comes at third.
APFD values among these three search strategies exhibit
slight differences, as all of them are inspired from one com-
mon search strategy, which is best first search strategy. Fire-
fly strategy for selecting next path solely depends on next
brightest agent. When there is one or more agents with equal
brightness, an agent will be randomly selected, resulting in
distinguishable best path in every iteration. Meanwhile, LBS
algorithm hardly depends on a fixed constant width value,
where different constant values set subsequently produce
different results. Highest APFD mean values for gzip and
flex programs in the experiments were also yielded by Firefly
Algorithm with values, 0.9501 and 0.9553, respectively while
for tcas benchmark program is 0.9298.
132370 VOLUME 7, 2019

TABLE 17. Summary OF APFD and time performance.
For overall time results, size of test cases does affect the
execution efficiency of prioritization process. The greater is
the number of test cases, the longer is the time taken to
complete an entire prioritization process. For tcas benchmark
program, which has the largest number of test cases, Firefly
strategy scored the best prioritization execution time, 390 sec-
onds in average, while LBS recorded 400 seconds, which
is the second fastest. In flex benchmark program, Firefly
and LBS recorded an equal execution time, 210 seconds.
However, in gzip program, LBS outperforms other algorithms
(180 seconds), five seconds much faster than Firefly in priori-
tizing test cases. This result could be attributed to the behavior
of LBS algorithm inspired by best search first concept which
possesses a superior execution time. Refined attractiveness
function that was used to determine firefly brightness has
managed to reduce the count of agents with equal brightness.
In turn, this has reduced the number of random selection of
agents when there are more agents with equal brightness.
Consequently, this reduced overall execution time for the
Firefly in prioritizing test cases. For LBS, the difference
in brightness values does have an influence on execution
time. LBS algorithm hardly depends on a constant value
for beam width, which inevitably has affected the overall
time execution. GA on the other hand, exhibited a unique
behavior with its adaptive search, where each solution path
produced is distinct attributed to its mutation fitness func-
tion at a considerably longer execution time. For Greedy
algorithm, the algorithm searches for agents which possess
highest brightness first, which may yield a sub-optimal solu-
tion path. Overall, Firefly Algorithm performs better in pri-
oritizing small to medium size test suites. FA performance
across three benchmark programs did not indicate stark
differences in terms of average execution time, indicating
its capability in yielding approximately acceptable solution
paths with a certain extent of consistency. In addition, FA also
produces better APFD measures indicating its suitability
to be used as a competitive prioritizing algorithm in TCP
research area.
VII. THREAT OF VALIDITY
This paper has some limitation acknowledged which may
possibly threaten its legitimacy. The potential threat of this
applied soft computing paper is associated with less diversity
in selection of benchmark program and incomplete data
extraction.
A. SELECTION OF DATASET
In benchmark program selected in this paper were only
four where three are from Real-world UNIX Benchmarks
Programs and three from Siemens Suites where all can be
obtained from http://sir.unl.edu [43]. From Table 5, the lim-
itation of Real-world UNIX Benchmarks Programs is, all of
them have small test cases pool. Possible issues may arise if
the proposed technique applied bigger sizer of test cases pool.
To reduce this problem, TCAS program were used as it has
large test cases pool. Other than selection of benchmark pro-
gram, authors does not apply the algorithm into the real cases
study which might have different kind of test cases structure
which may have significant affect towards the results.
B. INCOMPLETE DATA EXTRACTION
In this experiment, the data extracted from the test cases
were only the input data and also the output data. The value
for distances and weight of a test cases were solely based
on input data of each test cases while the output is used as
the fault matrix for APFD calculation. There are a lot of
other data were not fully utilized in this experiment such
as the system coding itself, the instruction within the test
cases, the steps and others. This incomplete data extraction
could be the next potential threat to this paper. To reduce this
problem, instead of using one calculation metric, authors used
two type of string metric which is string distance metric and
weight metric to assign the attractiveness to each firefly agent
accurately.
C. INCOMPLETE DATA EXTRACTION
In this paper, the experiment was solely focused on test-case
based TCP technique. Other techniques which have quite
similar strategies with this technique were not included in our
validation. It would be unfair if we compared different base of
TCP technique to assess which algorithm is better. To reduce
this problem, we had run the experiment for 30 times for each
algorithm to see the consistency of the result obtained. How-
ever, it might not be enough to see which algorithm actually
fits better in which TCP techniques. This is an opportunity
for future work to explore.
VIII. CONCLUSION
This paper used a Firefly strategy in prioritizing test cases
to achieve higher APFD results. Three benchmark programs
were used to validate the efficiency and effectiveness of
Firefly Algorithm in prioritizing test suites, in comparison
with other prioritization algorithms. The experimental results
showed that Firefly obtained the highest APFD scores com-
pared to other prioritization algorithms. The experiment also
showed that Firefly Algorithm slightly outperformed LBS in
terms of execution time. Overall, the APFD results showed
VOLUME 7, 2019 132371

that Firefly Algorithm may become a strong competitor in
the TCP area. The APFD results indicate that Firefly Algo-
rithm could be effective in discovering fault proneness issues,
which is strongly required in safety critical systems. In future
work, it would be interesting to explore possible enhancement
on this particular swarm intelligence algorithm focusing on
coverage efficiency.
ACKNOWLEDGMENT
The authors would like to express their sincere gratitude
to members of Embedded & Real- Time Software
Engineering Laboratory (EReTSEL), Software Engineer-
ing Research Group (SERG), Faculty of Engineering,
UTM and MIMOS for their feedback and continuous
support.
REFERENCES
[1] G. J. Myers, T. M. Thomas, and C. Sandler, The Art of Software Testing,
vol. 1. Hoboken, NJ, USA: Wiley, 2004.
[2] G. Rothermel, R. H. Untch, C. Chu, and M. J. Harrold, ‘‘Test case prioriti-
zation: An empirical study,’’ in Proc. IEEE Int. Conf. Softw. Maintenance
(ICSM), Aug./Sep. 1999, pp. 179–188.
[3] M. Khatibsyarbini, M. A. Isa, and D. N. A. Jawawi, ‘‘A hybrid weight-
based and string distances using particle swarm optimization for prioritiz-
ing test cases,’’ J. Theor. Appl. Inf. Technol., vol. 95, no. 12, pp. 2723–2732,
2017.
[4] M. Khatibsyarbini, M. A. Isa, D. N. A. Jawawi, and R. Tumeng, ‘‘Test
case prioritization approaches in regression testing: A systematic literature
review,’’ Inf. Softw. Technol., vol. 93, pp. 74–93, Jan. 2018.
[5] B. Jiang and W. K. Chan, ‘‘Input-based adaptive randomized test case
prioritization: A local beam search approach,’’ J. Syst. Softw., vol. 105,
pp. 91–106, Jul. 2015.
[6] W. E. Wong, J. R. Horgan, S. London, and H. Agrawal, ‘‘A study of
effective regression testing in practice,’’ in Proc. 8th Int. Symp. Softw. Rel.
Eng., Nov. 1997, pp. 264–274.
[7] G. Rothermel, R. H. Untch, C. Chu, and M. J. Harrold, ‘‘Prioritizing test
cases for regression testing,’’ IEEE Trans. Softw. Eng., vol. 27, no. 10,
pp. 929–948, Oct. 2001.
[8] S. Elbaum, A. G. Malishevsky, and G. Rothermel, ‘‘Test case prioritization:
A family of empirical studies,’’ IEEE Trans. Softw. Eng., vol. 28, no. 2,
pp. 159–182, Feb. 2002.
[9] Y. Singh, ‘‘Systematic literature review on regression test prioritization
techniques,’’ Informatica, vol. 36, pp. 379–408, Dec. 2012.
[10] Y. Ledru, A. Petrenko, S. Boroday, and N. Mandran, ‘‘Prioritizing test cases
with string distances,’’ Autom. Softw. Eng., vol. 19, no. 1, pp. 65–95, 2012.
[11] L. Mei, W. K. Chan, T. H. Tse, B. Jiang, and K. Zhai, ‘‘Preemptive
regression testingof workflow-based Web services,’’ IEEE Trans. Services
Comput., vol. 8, no. 5, pp. 740–754, Sep./Oct. 2015.
[12] S. W. Thomas, H. Hemmati, A. Hassan, and D. Blostein, ‘‘Static test case
prioritization using topic models,’’ Softw. Eng., vol. 19, no. 1, pp. 182–212,
2014.
[13] D. Gao, X. Guo, and L. Zhao, ‘‘Test case prioritization for regression
testing based on ant colony optimization,’’ in Proc. IEEE Int. Conf. Softw.
Eng. Service Sci., Sep. 2015, pp. 275–279.
[14] R. M. Parizi, A. Kasem, and A. Abdullah, ‘‘Towards gamification in soft-
ware traceability: Between test and code artifacts,’’ in Proc. 10th ICSOFT,
Jul. 2015, pp. 1–8.
[15] A. Shahbazi and J. Miller, ‘‘Black-box string test case generation through
a multi-objective optimization,’’ IEEE Trans. Softw. Eng., vol. 42, no. 4,
pp. 361–378, Apr. 2016.
[16] C. Catal and D. Mishra, ‘‘Test case prioritization: A systematic mapping
study,’’ Softw. Qual. J., vol. 21, no. 3, pp. 445–478, 2013.
[17] W. H. Gomaa and A. A. Fahmy, ‘‘A survey of text similarity approaches,’’
Int. J. Comput., vol. 68, no. 13, pp. 13–18, 2013.
[18] M. Tyagi and S. Malhotra, ‘‘Test case prioritization using multi objective
particle swarm optimizer,’’ in Proc. Int. Conf. Signal Propag. Comput.
Technol. (ICSPCT), Jul. 2014, pp. 390–395.
[19] P. R. Srivatsava, B. Mallikarjun, and X. S. Yang, ‘‘Optimal test sequence
generation using firefly algorithm,’’ Swarm Evol. Comput., vol. 8,
pp. 44–53, Feb. 2013.
[20] N. Iqbal, K. Zafar, and W. Zyad, ‘‘Multi-objective optimization of test
sequence generation using multi-objective firefly algorithm (MOFA),’’ in
Proc. Int. Conf. Robot. Emerg. Allied Technol. Eng. (iCREATE), Apr. 2014,
pp. 214–220.
[21] H. K. N. Leung and L. White, ‘‘Insights into regression testing,’’ in Proc.
Int. Conf. Softw. Maintenance, Oct. 1989, pp. 60–69.
[22] S. Elbaum, G. Rothermel, and J. Penix, ‘‘Techniques for improving
regression testing in continuous integration development environments,’’
in Proc. 22nd ACM SIGSOFT Int. Symp. Found. Softw. Eng. (FSE), 2014,
pp. 235–245.
[23] P. K. Chittimalli and M. J. Harrold, ‘‘Recomputing coverage information
to assist regression testing,’’ IEEE Trans. Softw. Eng., vol. 35, no. 4,
pp. 452–469, Jul. 2009.
[24] S. Yoo and M. Harman, ‘‘Regression testing minimization, selection and
prioritization: A survey,’’ Softw., Test. Verification Rel., vol. 22, no. 2,
pp. 67–120, 2012.
[25] H. Hemmati, A. Arcuri, and L. Briand, ‘‘Empirical investigation of the
effects of test suite properties on similarity-based test case selection,’’ in
Proc. 4th IEEE Int. Conf. Softw. Test., Verification Validation, Mar. 2011,
pp. 327–336.
[26] R. Maheswari and D. Mala, ‘‘Combined genetic and simulated annealing
approach for test case prioritization,’’ Indian J. Sci. Technol., vol. 8, no. 35,
pp. 1–5, 2015.
[27] Y. Lou, D. Hao, and L. Zhang, ‘‘Mutation-based test-case prioritization
in software evolution,’’ in Proc. IEEE 26th Int. Symp. Softw. Rel. Eng.
(ISSRE), Nov. 2015, pp. 46–57.
[28] F. Yuan, Y. Bian, Z. Li, and R. Zhao, ‘‘Epistatic genetic algorithm for test
case prioritization,’’ in Proc. Int. Symp. Search Based Softw. Eng., 2015,
pp. 109–124.
[29] C. Catal, ‘‘On the application of genetic algorithms for test case prioritiza-
tion: A systematic literature review,’’ in Proc. 2nd Int. Workshop Evidential
Assessment Softw. Technol., 2012, pp. 9–14.
[30] A. Kaur and S. Goyal, ‘‘A genetic algorithm for fault based regression test
case prioritization,’’ Int. J. Comput. Appl., vol. 32, no. 8, pp. 975–8887,
2011.
[31] W. Jun, Z. Yan, and J. Chen, ‘‘Test case prioritization technique based
on genetic algorithm,’’ in Proc. Int. Conf. Internet Comput. Inf. Services,
Sep. 2011, pp. 173–175.
[32] S. Sabharwal, R. Sibal, and C. Sharma, ‘‘Prioritization of test case scenar-
ios derived from activity diagram using genetic algorithm,’’ in Proc. Int.
Conf. Comput. Commun. Technol. (ICCCT), Sep. 2010, pp. 481–485.
[33] K. Deb, S. Pratap, S. Agarwal, and T. Meyarivan, ‘‘A fast and elitist
multiobjective genetic algorithm: NSGA-II,’’ IEEE Trans. Evol. Comput.,
vol. 6, no. 2, pp. 182–197, Apr. 2002.
[34] H. Do and G. Rothermel, ‘‘On the use of mutation faults in empirical
assessments of test case prioritization techniques,’’ IEEE Trans. Softw.
Eng., vol. 32, no. 9, pp. 733–752, Sep. 2006.
[35] Z. Li, M. Harman, and R. M. Hierons, ‘‘Search algorithms for regres-
sion test case prioritization,’’ IEEE Trans. Softw. Eng., vol. 33, no. 4,
pp. 225–237, Apr. 2007.
[36] S. Li, N. Bian, Z. Chen, D. You, and Y. He, ‘‘A simulation study on some
search algorithms for regression test case prioritization,’’ in Proc. 10th Int.
Conf. Qual. Softw., Jul. 2010, pp. 72–81.
[37] A. K. Joseph, G. Radhamani, and V. Kallimani, ‘‘ Improving test efficiency
through multiple criteria coverage based test case prioritization using
Modified heuristic algorithm,’’ in Proc. 3rd Int. Conf. Comput. Inf. Sci.
(ICCOINS), Aug. 2016, pp. 430–435.
[38] S. Eghbali and L. Tahvildari, ‘‘Test case prioritization using lexicograph-
ical ordering,’’ IEEE Trans. Softw. Eng., vol. 42, no. 12, pp. 1178–1195,
Dec. 2016.
[39] K. Solanki, Y. Singh, S. Dalal, and P. R. Srivastava, ‘‘Test case pri-
oritization: An approach based on modified ant colony optimization,’’
in Emerging Research in Computing, Information, Communication and
Applications. 2016.
[40] T. Noguchi, H. Washizaki, Y. Fukazawa, A. Sato, and K. Ota, ‘‘History-
based test case prioritization for black box testing using ant colony opti-
mization,’’ in Proc. IEEE 8th Int. Conf. Softw. Test., Verification Validation
(ICST), Apr. 2015, pp. 1–2.
[41] P. Stratis and A. Rajan, ‘‘Test case permutation to improve execution time,’’
in Proc. 31st IEEE/ACM Int. Conf. Automat. Softw. Eng. (ASE), 2016,
pp. 45–50.
132372 VOLUME 7, 2019

[42] A. Aizawa, ‘‘An information-theoretic perspective of tf–idf measures,’’ Inf.
Process. Manage., vol. 39, no. 1, pp. 45–65, Jan. 2003.
[43] Software-Artifact Infrastructure Repository: Home. [Online]. Available:
http://sir.unl.edu/portal/index.php
[44] L. Zhang, D. Hao, L. Zhang, G. Rothermel, and H. Mei, ‘‘Bridging the
gap between the total and additional test-case prioritization strategies,’’ in
Proc. Int. Conf. Softw. Eng., 2013, pp. 192–201.
MUHAMMAD KHATIBSYARBINI was born in
Selama, Perak, Malaysia. He received the B.S.
degree in software engineering from the Universiti
Teknologi Malaysia, Johor, Malaysia, in 2016, and
the M.S. degree in master of philosophy from the
Universiti Teknologi Malaysia, in 2018. During
his study, he was a Research Assistant (2016–
2018) with the Embedded and Real-Time Software
Engineering Laboratory (EReTSEL) and Software
Engineering Research Group (SERG). Also, dur-
ing those two years, he has been working as a contract Software Engineer
with two different companies: Mechabotic Enterprise and Effron Sdn Bhd.
His research interests include software engineering, search-based software
testing, and artificial intelligence.
MOHD ADHAM ISA received the bachelor’s
degree in computer science from Universiti
Teknologi Malaysia, the master’s degree in com-
puter science, and the Ph.D. degree in software
engineering from Universiti Teknologi Malaysia
(UTM), Malaysia, where he is currently the Head
of the Software Engineering Research Group
(SERG). His main research interests include soft-
ware engineering, software quality, software test-
ing, requirement engineering, and software project
management. A major part of his research projects focuses on software
quality assurance, real-time embedded systems, as well as the Internet of
Things (IoT).
DAYANG N. A. JAWAWI received the bachelor’s
degree in software engineering from Sheffield Hal-
lam University, U.K., and the master’s degree in
computer science and the Ph.D. degree in software
engineering from Universiti Teknologi Malaysia
(UTM), Malaysia, where she is currently an Asso-
ciate Professor with the School of Computing, Fac-
ulty of Engineering. Her main research interests
include software engineering, software reuse, soft-
ware quality, software testing, requirement engi-
neering, and computing education. A major part of her research projects
focuses on rehabilitation and mobile robotics, real-time embedded systems,
as well as precision farming applications.
HAZA NUZLY ABDULL HAMED received the
first degree in information technology, majoring
in artificial intelligence, from Universiti Utara
Malaysia, the master’s degree in computer science
from Universiti Teknologi Malaysia, and the Ph.D.
degree from the Auckland University of Technol-
ogy, New Zealand. He is currently a Senior Lec-
turer with the School of Computing and a found-
ing member of the Applied Industrial Analytics
Research Group (ALIAS), Faculty of Engineering,
Universiti Teknologi Malaysia (UTM). Before joining UTM, he worked as
a Web Programmer and System Analyst. His research interests are compu-
tational intelligence, evolutionary computation, deep learning, optimization,
spatiotemporal data processing, and information system development.
MUHAMMAD DHIAUDDIN MOHAMED SUF-
FIAN received the B.Tech. (Hons.) degree in busi-
ness information systems and the M.Sc. degree in
computer science – real-time software engineer-
ing from Universiti Teknologi PETRONAS and
Universiti Teknologi Malaysia, respectively. He is
currently the Senior Staff Engineer with the Busi-
ness Solutions and Services Division, MIMOS
Technology Solutions Sdn. Bhd, a subsidiary of
MIMOS Berhad, the National Applied R&D Cen-
tre. He is a certified Six Sigma Green Belt, a Certified Tester Advanced
Level-Test Manager (CTAL-TM), and a Certified Tester Foundation Level
(CTFL). His experiences encompass various spectrums of ICT across various
sectors, including government, automotive, banking, and education. Cur-
rently, he assumes the role of Lead Business System Analyst for an ICT Court
Transformation Project. His research interests focus on software require-
ment, software testing, software quality, and software process. He holds two
patents filed under his name, of which one as main inventor and another one
as co-inventor.
VOLUME 7, 2019 132373

Test case prioritization using firefly algorithm for software testing

Recommended

More Related Content

What's hot (20)

Similar to Test case prioritization using firefly algorithm for software testing (20)

More from Journal Papers (20)

Recently uploaded (20)

Test case prioritization using firefly algorithm for software testing