IOP Conference Series: Materials Science and Engineering

PAPER • OPEN ACCESS

A Self-Adapting Ant Colony Optimization Algorithm Using Fuzzy Logic

(ACOF) for Combinatorial Test Suite Generation
To cite this article: Mohd Zamri Zahir Ahmad et al 2020 IOP Conf. Ser.: Mater. Sci. Eng. 767 012017

View the article online for updates and enhancements.

This content was downloaded from IP address 178.171.114.220 on 21/03/2020 at 17:15

1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

A Self-Adapting Ant Colony Optimization Algorithm Using

Fuzzy Logic (ACOF) for Combinatorial Test Suite Generation

Mohd Zamri Zahir Ahmad, Rozmie Razif Othman, Mohd Shaiful Aziz Rashid
Ali and Nuraminah Ramli.
School of Computer and Communication Engineering, Kampus Pauh Putra, Universiti
Malaysia Perlis, 02600 Arau, Perlis, Malaysia.

E-mail: [email protected], [email protected],

[email protected], [email protected].

Abstract. Software testing is one of most crucial phase in software development life
cycle (SDLC). The main function of testing is to cater bugs between interactions of the
inputs. It is not possible to eliminate all bugs in one system but by using a suitable
testing optimization, it can provide a good enough solution for it. Reducing effort on
this phase is not only could lead to numerous bugs between the input interactions, but it
also leads to a greater loss such as loss of profits, reputations and even loss of life.
Generally, there are three categories of combinatorial testing techniques which is
computational, metaheuristic, and hyper heuristic. Ant colony optimization (ACO) is a
heuristic technique where its mimic the nature of ants in finding the best route from the
nest to the food node and vice versa. Many optimization problems have been solved by
using ACO. This paper is to proposed a self-adapting ant colony optimization algorithm
using fuzzy logic (ACOF) for combinatorial test suite generation, where it will
dynamically determine number of ants and edge selection (i.e. either to explore or to
exploit) based on percentage of remaining tuple list and covered test cases.

Keywords: Combinatorial testing, T-way testing, Ant Colony Optimization.

1. Introduction
In development of a software system, there are five main stages that engage throughout of the entire
process. Also knows as software development life cycle (SDLC), it consists of software requirement
analysis, system design, implementation, testing and evolution. It is reported that, from this five stages,
testing phase consume up to 50% of total cost development. This numbers could growth if the system
involves with safety-critical applications [1]. Software failure is a disaster situation as it could increase
the cost of software development tremendously. In order to minimize the software failure, a rigorous
testing needs to be conducted. To ensure the effectiveness of the testing activity, many software testing
technique have been developed in literature. Combinatorial testing (or also known as t-way testing) is
one of the test design technique that has been used widely. It can be used to generate test suite that can
cater bugs due to interaction between input parameters [2]. In generating the t-way test suite, t-way
strategy uses either mathematical or computational methods in reducing the test suite size but at the
same time still covering all interactions involving specified parameters-values combination for specific
strength, t [3,4].

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

As the main aim of all combinatorial test suite generation strategy is to produce as small test suite
size as possible, many optimization algorithm has been proposed and adopted for the purpose of
generating t-way test suite (e.g IPOG, Jenny, GTWay, ITTDG and TVG). At the same time, there are
researchers that use metaheuristic optimization technique in generating t-way test suite. Bestoun
implements Particle Swarm Optimization technique in generating uniform and variable strength t-way
test suite while Alsewari adopted Harmonic Search Algorithm for generating uniform strength t-way
test suite. There is also an attempt made by Zamli of using more than one metaheuristic technique
(namely hyper-heuristic technique) to generate uniform strength t-way test suite. The capability of
metaheuristic techniques in producing smaller size of test suite become the motivation of researchers to
implements these techniques in generating t-way test suite.
One of the most used metaheuristic technique in generating t-way test suite is Ant Colony
Optimization (ACO) algorithm. In 2004, Shiba successfully implements ACO algorithm in generating
uniform strength t-way test suite [5]. Enhancing Shiba works, Chen added local search in his
implementation of ACO (called Ant Colony System) for generating variable strength t-way test suite
[6]. Recently, Ramli proposed her implementation of ACO algorithm for input-output based t-way test
suite [7]. Overall, all implementation of ACO successfully producing the smallest t-way test suite in
many system configurations [5,6,8–12].
The success of ACO in producing smaller test suite size is highly depending on number of ants used
in the algorithm. Increasing number of ants increase the probability of producing the optimum test suite.
Nevertheless, increasing number of ants will also increase the execution time. Although the execution
time are also affected by system environment used, optimizing the numbers of ant possibly could
optimize the ant colony efficiency. Thus, we propose a study on a self-adapting ACO algorithm using
fuzzy logic for combinatorial test suite generation. The fuzzy component will dynamically determine
the suitable number of ants that will be used in ACOF and edge selection either to explore or to exploit
the search space.

2. Related Works
Particle Swarm Optimization (PSO) is an algorithm for t-way combinatorial testing published in 2010.
The algorithm is inspired by swarm animals hunting for their food. Each individual in the swarm moves
towards the best individual location and the best global location based on optimal solution calculated
from their position and velocity [13]. A few algorithm or tools is developed based on PSO algorithm.
The generation of PSO algorithm for t-way testing starts with the development of pairwise PSO [13].
The researchers have developed two different algorithms. One algorithm is OTAT strategy and another
one is OPAT. The utilization of PSO algorithm in the previous research has encouraged the development
of Particle Swarm Test Generator (PSTG) [9]. Faster convergence rate attitude, requires a few
parameters to be controlled, easily apply to any optimization problem and lighter computational load
are the reasons PSO has been chosen as PSTG’s basic platform [9,10]. PSTG only support uniform
strength up to 6 levels. Next, VSPSTG has been developed to improve PSTG by enhancing variable
strength interaction [11]. Discrete Particle Swarm Optimization (DPSO) [14] is another algorithm based
on PSO and discrete particle swarm as its basis. This new algorithm is designed by adopting S-PSO that
uses set-based scheme as its discrete search space. The new DPSO has improves the performance by
having two auxiliary strategies; particle reinitialization and additional evaluation of gbest. It also offers
guidelines for parameter settings. DPSO is reported to be a promising improvement of PSO. The last
algorithm that motivated by PSO between 2010 and 2017 is Swarm Intelligent Test Generator (SITG)
[12]. SITG could support 2 to 6 uniform strength and variable strength. SITG outperform PSTG in some
cases in uniform strength while performing better results for variable strength. As PSO is a metaheuristic
algorithm, its entire variant follows the similar search technique. PSTG, V-PSTG, DPSO and SITG use
OTAT strategy approach while Pairwise PSO has both OTAT and OPAT strategy for different type of
Pairwise PSO algorithm.
In 2011, GTWay, ITTDG and AURA were introduced as another t-way combinatorial testing
algorithms. Those algorithms applied computational search technique to find the best test cases.

2
1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

GTWay [15] applies backtracking concept. The algorithm needs to map the actual data with symbolic
representation. Then, it generates t-way pair possible interactions. After that, the strategy backtracks the
uncovered t-way pairs. The t-way pairs are merge if they are combinable and complement each other’s
missing value and the new merge interaction could cover the most uncovered t-way pairs. If the pairs
fail to merge, backtrack algorithm goes back to the first defined values. GTWay is an OTAT strategy
that supports uniform strength for t greater than six. ITTDG [8] practices a list of candidate test data to
be used in deciding a final test data. The visited tuples are put into a candidate list. Next, parameter that
has a value that covers the most uncovered tuples is added to the candidate test data, one parameter at a
time. In a situation where tie occurs, the corresponding test data that has the tie values is duplicated and
put into the candidates list. Once the candidate list finished forming a complete test data, the strategy
calculates weight for every test data. Test data with a highest weight has a value of the most covered
tuples. If tie happens, the first round in the candidate list will be selected to be put in the final test suite
and removed from the uncovered tuples list. The process repeated until all uncovered tuples are being
covered. ITTDG could support all three types of interaction; uniform, variable and IOR.
Similar to ITTDG, AURA [16] has the same ability in supporting all type of interactions. AURA
started by exploiting the interactions combination created earlier to generate test suite. Lastly actual data
mapping algorithm is used to support symbolic values and actual data output. Only one type of algorithm
published in 2012. Harmony Search algorithm is the basis for two other algorithms, HSS and HS-
PTSGT. Another algorithm that uses metaheuristic search technique to generate t-way test suite is
Harmony Search Strategy (HSS) [4,15]. It is based on Harmony Search (HS) algorithm. Similar to PSO,
HS algorithm requires lightweight computation with only few parameters setting. HSS could support
uniform and variable interaction strength with high interaction strength, up to 14 and constraints. Besides
HSS, HS-PTSGT [6] is a pairwise generator tool that was developed based on HS algorithm.
In 2013 DA-RO and DA-FO [17] emerged. Both algorithms are a computational technique and
support all type of interactions. DA-FO starts with an empty test case which contains unfixed factors.
Local density for each factor is determined to get the order of coverage requirements. A coverage
requirement with highest local density is selected to be fixed factor based on global density. For each
combination in the test case where the values of factors are fixed, DA-RO will calculate global density.
The combination which holds greatest global density is selected to fix the value of factors. Unlike DA-
RO, DA-FO algorithm produces single test case by fixing value in order of factor. Priority number is
defined to measure priority of different factors to determine order of factors. Coverage requirements
that hold high priorities and local densities will get a great factor density and will be chosen to fix the
values. Similar to DA-RO, after fixing values, the local, global and factor densities may change due to
re-modification of densities. DA-FO also suffers from tie problem. To break the ties, the similar factors
as DA-RO can be used. GVS [18] is the only algorithm published in 2014. The algorithm is inspired by
GTWay. Similar to GTway, GVS is a computational technique and improves the interaction by
supporting all type of interactions. It generates one tuple at a time before the iteration starts. Each tuple
which is generated in the tuple generator is assigned with a don’t care value to complete the test data.
The test data with the first don’t care value, is compared with the uncovered tuples list. If the test data
exists, the test data with the first don’t care is selected. However, if not, the algorithm changes the next
remaining value parameter in the don’t care value space. The process of determining the don’t care
values is repeated until all parameter values finish. Afterward, GVS checks which test cases could cover
the most uncovered tuples. The chosen test case is then placed into final test suite and covered tuples
are added in the covered tuples list.
In 2015, TCA, Cuckoo Search (CS) and Flower Strategy (FS) were published. They are metaheuristic
algorithms. TCA [19] is the algorithm that combines greedy Tabu search and random walk heuristic.
Greedy Tabu search is used during initialization of test cases, and then TCA apply heuristic search
technique to expand the opportunities in covering the uncovered interactions. By using both techniques,
it effects the runtime test suite generation. Even though TCA utilizes combination of search techniques,
it is still under metaheuristic category because Tabu search is a metaheuristic algorithm. TCA supports
3-way constraint uniform strength. Cuckoo Search [20] has been explored in t-way combinatorial testing

3
1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

as a capacity that Cuckoo algorithm offers search capabilities using Levi flights [21] to update search
space and it only consists of few parameters to be tuned [20]. Cuckoo Search supports uniform, up to 3
interaction strength and also variable strength. Flower Strategy (FS) [22] evolved from the beauty of
Flower Pollination Algorithm (FPA). FPA is a simple, flexible and requires lightweight computation. It
also offers a balance between exploitation and exploration by employing levi flight. FS is a metaheuristic
that supports uniform strength.
In 2016, hyper-heuristic strategy emerged. High Level Hyperheuristic (HHH) [23] is the pioneer in
utilizing hyper-heuristic strategy in t-way combinatorial testing. It consists of high level metaheuristic
Tabu Search and other four (4) low level metaheuristic algorithms (i.e. Teaching Learning based
Optimization, Global Neighbourhood Algorithm, Particle Swarm Optimization, and Cuckoo Search
Algorithm). The algorithm uses metaheuristic search technique and could support uniform strength from
2 until 6 level of strength. There are two proposed t-way testing strategies found in the literature.
Artificial Bee Colony (ABC) [24] has been proposed to generate an optimum test suite. The strategy is
inspired by a group of bees searching for nectar for their hive. ABC was chosen by the researchers as it
was proven to be a great strategy for combinatorial field. Another proposed strategy found in 2016 by
Ramli is Ant Colony System (ACS) [7]. ACS is a variant of Ant Colony Optimization (ACO) algorithm.
It has been successfully solving many combinatorial optimization problems. Ant Colony Algorithm
(ACA) [5] and ACS [6] have been built to generate an optimum test suite to cater uniform and variable
strength respectively. The proposed ACS strategy is to cater all type of interactions especially IOR.
Adaptive TLBO (ATLBO) is a metaheuristic t-way testing that emerged from Teaching Learning-based
optimization (TLBO) algorithm in the first quarter of 2017. It was based on Mamdani fuzzy inference
system. ATLBO could support uniform and variable strength [25].

3. Problem Definition Model

To demonstrate uniform strength of t-way testing, let’s consider a system for scholarship sponsor
application. The system is to help filtering application from student to determine which student is
eligible to obtain the scholarship. There system will filter the application based on four factor which is
cumulative grade point average (CGPA), course, study level and parent salary as shown in Table 1. To
ease the discussion, a representation of parameters and its value is shown in Table 2. The CGPA is
represented as ‘C’, course represented as ‘K’, study level represented as ‘S’ and parent salary
represented as ‘P’. For the values, each of the values represented by small letter of the represented
filtering factor. For example, c1 referring to input parameter CGPA and the value is less and equal to 3.

Table 1. Requirement for Scholarship Application.

Filtering Factor CGPA Course Study Level Parent Salary
Values ≤3 Engineering Diploma ≤ RM2000
>3 Business Bachelor Degree > RM2000

Table 2. Simplified Inputs Parameter for Scholarship Application.

Input Parameter C K S P
c1 k1 s1 p1
Values c2 k2 s2 p2

To minimize the potential bugs in this system, appropriate testing must be made before release the
system to end user. Interaction testing is the crucial test that need to be conduct as fault may occurred
between this input parameter. One of the technique to test the system is by using exhaustive testing. In
exhaustive testing, all possible input combinations need to be tested. This means, all input is interacting
with each other where faults are expected to arise between the interactions. As a result, 16 test cases (i.e.
24 = 16) need to be executed. The list of exhaustive test case is shown in Table 3.

4
1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

Table 3. Exhaustive Test Suite for Scholarships Application.

C K S P
c1 k1 s1 p1
c1 k1 s1 p2
c1 k1 s2 p1
c1 k1 s2 p2
c1 k2 s1 p1
c1 k2 s1 p2
c1 k2 s2 p1
c1 k2 s2 p2
c2 k1 s1 p1
c2 k1 s1 p2
c2 k1 s2 p1
c2 k1 s2 p2
c2 k2 s1 p1
c2 k2 s1 p2
c2 k2 s2 p1
c2 k2 s2 p2

Table 3 shown exhaustive testing of the system where the strength, t is 4. The interaction strength is
possible to be reduce to 2 (t=2) or also known as pairwise testing. This means, only two input parameter
are interacting and faults is expected to arise within these interactions as compared to four input
parameter in the exhaustive testing. The rest of input parameter that not related to interaction are
considered as ‘don’t care’ (DC) value, as it will not affect the input interactions testing. The purpose to
reduce the strength is to minimize the generated test case without compromise the interacting input. The
possible input interaction for 2-way testing is CK, CS, CP, KS, KP and SP. As the result, the generated
test case now become smaller where the repeated test case is removed from the list as shown in Figure
1. Noted that, it almost achieves 44 percent reduction of test cases as compared to exhaustive testing.
The reduction may not significant for a small and simple application, but it is significantly reduced for
a larger and complex applications environment. For example, system with 40 input with 2 possible value
each will resulted 1099511627776 (i.e. 240) test cases need to be executed. It is impractical to do the
exhaustive testing as it will consume a lot of time as well as many software testers to be part in, therefore
the overall cost of testing will increase enormously. Hence, by implementing t-way testing technique,
it is proven could reduce the test cases and save a lot of resources with an acceptable fault detection.

Final Test Suite

Figure 1. Test Suite for 2-Way Testing.

5
1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

4. Overview of ACO
Ant algorithm is a metaheuristic approach where it is inspired by nature life behavior of ant colony in
finding shortest path to foods from their nest. Some ants’ species are completely blind and they can’t
see anything, yet they have a very powerful searching system where they can find shortest path to the
foods source. Ants sense the path by a special chemical deposited by them while travelling the path
called as pheromone. The more pheromone deposited to the trail increases the probability for the ants to
choose that path.
There were several algorithms have been proposed inspired from the nature behavior of ant colony.
In early nineties, an Ant System (AS) was the first proposed by Dorigo et al. [26] for travelling salesman
problem (TSP). Ant system consist of three main techniques which is ant-cycle, ant-density, and ant-
quantity. Among of this three techniques, ant-cycle is the most performed compared the other two hence
it was later established as the Ant System [27]. This algorithm was then being enhance producing several
more algorithm based on this Ant System. Among of the rise algorithm are Ant System Elitist, rank-
based Ant System (ASrank), MAX–MIN Ant System (MMAS), and Ant Colony System (ACS). ACS
is the tuned version of AS where several modifications been made to enhance the searching
effectiveness. It is reported that ACS give the most competitive performance compared with others AS
strategies because it goods in the exploitation and exploration in searching of the best solutions [6,28].
To be comprehensive, ACS is improved because of;
a) Edge selection
The edge selection rule provides, which called as random-proportional rule; a through probability
method to balance searching solutions, between exploration of the new path or exploitation of the
path with previous information. A random variable of q and user-define 𝑞0 is used to define the
probability of the ants in selecting the edges. Variable q is uniformly distributed in [0,1] and variable
𝑞0 (0 ≤ 𝑞0 ≤ 1) is used to determine either the ant to do the exploration or exploitation.
 Exploration; if (𝑞0 > 𝑞)
𝛼 𝛽
[𝜏𝑖,𝑗 (𝑡)] [𝜂𝑖,𝑗 (𝑡)]
𝑝𝑖,𝑗 (𝑡) = 𝛼 𝛽
(1)
𝑙
∑ℎ=1
𝑖
[𝜏𝑖,ℎ (𝑡)] [𝜂𝑖,ℎ (𝑡)]

 Exploitation; if (𝑞0 ≤ 𝑞)

𝛼 𝛽
𝐴𝑟𝑔𝑚𝑎𝑥1≤ℎ≤𝑙𝑖 {[𝜏𝑖,ℎ (𝑡)] [𝜂𝑖,ℎ (𝑡)] } (2)

where;
 [𝑡𝑖,𝑗 (𝑡)] is the amount of pheromone values of edges 𝑒𝑖,𝑗 ,
 𝜂𝑖,𝑗 (𝑡) is the heuristic values value of edge 𝑒𝑖,𝑗 , and
 α(α≤0) and β(β>0) are the factor that determine relative influence of pheromone and
heuristic value on the edge decision of the ant.
b) The pheromone update
 Global pheromone update
Ants that travel along the path will construct the path based on probability of the edge chosen.
Eventually there is only one best ant identified as the single best ant (or globally best ant) where
it constructs the shortest path compare to the others. This single best ant is the only ant that
allowed to deposit pheromone to the trail. Meanwhile, pheromone evaporation occurs in order
to avoid the algorithm stuck in a local minima solution. This is process done after all the ants
have successfully complete construct the path in the current iteration.

𝑏𝑠
(1 − 𝜌)𝜏𝑖,𝑗 + 𝜌Δ𝜏𝑖,𝑗 𝑖𝑓 𝑒𝑖,𝑗 ∈ 𝑡𝑒𝑠𝑡 𝑏𝑠
𝜏𝑖,𝑗 (𝑡 + 1) = { (3)
𝜏𝑖,𝑗 (𝑡) 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

6
1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

where;
 ρ[0,1] is the pheromone evaporation rate and,
 Δτ_(i,j)^bs is the fitness function.
 The local pheromone update
The main intention of local pheromone update is to make the search performed by
following ants during the iteration is diversified. While travelling in trail path, each ant
is allowed to deposit their pheromone at the last edges. Meanwhile, pheromone
evaporation occurs as to make sure subsequent ant explore the different edges. Ants will
update the pheromone level by applying local update rule;

𝜏𝑖,𝑗 (𝑡) = (1 − 𝜑)𝜏𝑖,𝑗 (𝑡) + 𝜑𝜏0 (4)

where;
 𝜏0 is the initial value of pheromone and
 𝜑 (0 < 𝜑 < 1) is the evaporation of the pheromone.

5. Proposed Strategy
The aim of t-way testing test is to find faults between interaction input parameter. In previous section,
the algorithm of ant and variant of ant colony optimization is explained. However, there is room for
improvement that still can be made to the algorithm, where number of ants is still not optimized. As
ACOF will adapt OTAT searching technique, number of ant’s agent can be optimized based on
remaining uncovered tuple using fuzzy logic algorithm. Besides, the probability in selecting the edge to
travers also need to be tuned with exploration in early iteration and exploitation in middle or near end.
Figure 2 shows the main component function of ACOF. It consists of four main components that drive
towards the effectiveness of ACOF.

Figure 2. Main Framework of ACOF

5.1. Ant Colony Path Generator (ACPG)

This component function is to generate path of the ant based on configuration setting set by software
tester. It consists of virtual nodes where it represents the inputs and branch for each node where it
represents value of each input. This path will be used by ant agents to travel from each input node, in
finding the optimum path. This will help in generating the optimum test case.

7
1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

5.2. Tuple List Generator (TLG)

This function is to generate a full list of interaction tuples. This functions retrieve inputs parameter (i.e.
number of input and its values), interaction parameters as well as interaction strength prior generating
the tuple list. The tuple list then will be used by test case generator to determine which test case covered
the most tuple. Tuple list will eventually will be empty as every covered tuple by test cases will be
removed from the list.

5.3. Ant Agent Fuzzy Controller (AAFC)

Different input configuration tends to have different amount of ant agent to be released. This function
is to dynamically assign number of ant agent without regarding the input setting. Different from others
ACO algorithm, number of ant agent will be optimized automatically using fuzzy logic algorithm. It is
based on the percentage of uncovered tuples in the tuple list and percentage of covered test case of every
iteration. For the early iteration, less number of ant will be assigned as there are still many uncovered
tuples in the list. While the iterations reach to half or near end, more ant will be assigned as to help test
case generator to find the most uncovered tuple.

5.4. Test Case Generator (TCG)

TCG is the main component to generate test case. As ACOF will adapt OTAT, one best test case will
be generated in one process iteration. Test case generator uses ACPG as the path for ants to travers in
constructing the test case. In traveling from node to node, ant will have to make decision in which branch
they will choose. The decision is based on heuristic value, pheromone, probability and fitness function.
However, as mentioned in AAFC earlier, the probability need to be tune as to control the exploration
and exploitation of the path travel by the ants. In finding the best test case for early iteration, ant will
explore the search space while in middle or near end of the process, ants will exploit path from the last
best test case. This is since, at beginning of the iteration, there are still a lot of uncovered tuple remain
in the tuple list. Thus it is to avoid the algorithm stuck in local minima solution, where there might be
other best solution in other space. Eventually, ant will do the exploitation as remaining of uncovered
tuple now become fewer to empty.

5.5. Flowchart of ACOF

An overview of how the strategy works is presented in Figure 3. The flowchart presents the process in
general. The strategy accepts input parameters and its values as input to the strategy and produces test
suite. Firstly, TLG generates a tuple list. In the meantime, ACPG starts to generate path to be travelled
by ant’s agent based on input parameters and its values. After both TLG and ACPG finished their tasks,
TCG starts to generate a single best test case.
Next, the generated single best test case will construct its own tuples. Each tuple then is compared
to the tuples list. Any similar tuples will be removed from the tuple list. The similar tuples mean that
the single best test case covers at least one combination of input parameters. The single best test case is
then stored in the Test Suite. However, if none of the constructed tuple is similar to the tuples in the
tuple list, the TCG continues to generate another single best test. Throughout the process, the tuple list
is becoming smaller as it has been removed whenever it is covered by a single best test. The process
iterates until the tuple list is empty.

8
1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

Figure 3. Flowchart of ACOF

6. Conclusion and Future Works

In this paper, we proposed a t-way strategy that based on ant colony optimization with fuzzy logic. The
framework of ACOF is consist of four main components which is Tuple Generator, Ant Colony Path
Generator, Ant Agent Controller and Test Case Generator. Next, the algorithm will be implemented and
the performance of the ACOF (i.e. test suite size) will be evaluated and compare against other strategies
with several benchmarking experiment that have been published.

Acknowledgement
The author would like to acknowledge the support from Fundamental Research Grant Scheme (FRGS)
under a grant number of FRGS/1/2018/ICT01/UNIMAP/02/1 from the Ministry of Education Malaysia.

References
[1] Ammann P and Offutt J 2008 Introduction to Software Engineering (Cambridge University
Press)
[2] Ahmad M Z Z, Othman R R and Rashid Ali M S A 2016 Sequence Covering Array Generator
(SCAT) for Sequence Based Combinatorial Testing Int. J. Appl. Eng. Res. 11 5984–91
[3] Kuhn D R, Kacker R N and Lei Y 2009 Combinatorial Software Testing Computers 42 94–6
[4] Alsewari A R A and Zamli K Z 2011 Interaction Test Data Generation Using Harmony Search
Algorithm 2011 IEEE Symp. Ind. Electron. Appl. ISIEA 2011 559–64
[5] Shiba T, Tsuchiya T and Kikuno T 2004 Using Artificial Life Techniques to Generate Test Cases
for Combinatorial Testing Proc. 28th Annu. Int. Comput. Softw. Appl. Conf. 2004. COMPSAC
2004. 72–7
[6] Chen X, Gu Q, Li A and Chen D 2009 Variable Strength Interaction Testing with an Ant Colony
System Approach Proc. - Asia-Pacific Softw. Eng. Conf. APSEC 160–7
[7] Ramli N, Othman R R and Ali M S A R 2017 Optimizing Combinatorial Input-Output Based
Relations Testing Using Ant Colony Algorithm 2016 3rd Int. Conf. Electron. Des. ICED 2016

9
1st International Symposium on Engineering and Technology (ISETech) 2019 IOP Publishing
IOP Conf. Series: Materials Science and Engineering 767 (2020) 012017 doi:10.1088/1757-899X/767/1/012017

586–90
[8] Othman R R and Zamli K Z 2011 ITTDG: Integrated T-way Test Data Generation Strategy for
Interaction Testing Sci. Res. Essays 6 3638–48
[9] Ahmed B S and Zamli K Z 2010 PSTG: A t-way strategy adopting particle Swarm Optimization
AMS2010 Asia Model. Symp. 2010 - 4th Int. Conf. Math. Model. Comput. Simul. 1–5
[10] Ahmed B S, Zamli K Z and Lim C P 2012 Application of Particle Swarm Optimization to
uniform and variable strength covering array construction Appl. Soft Comput. J. 12 1330–47
[11] Ahmed B S and Zamli K Z 2011 A Variable Strength Interaction Test Suites Generation Strategy
Using Particle Swarm Optimization J. Syst. Softw. 84 2171–85
[12] Rabbi K, Mamun Q and Islam R 2015 An Efficient Particle Swarm Intelligence Based Strategy
to Generate Optimum Test Data in T-way Testing 2015 IEEE 10th Conf. Ind. Electron. Appl.
123–8
[13] Chen X, Gu Q, Qi J and Chen D 2010 Applying Particle Swarm Optimization to Pairwise Testing
Proc. - Int. Comput. Softw. Appl. Conf. 107–16
[14] Wu H, Nie C, Kuo F C, Leung H and Colbourn C J 2015 A Discrete Particle Swarm Optimization
for Covering Array Generation IEEE Trans. Evol. Comput. 19 575–91
[15] Zamli K Z, Klaib M F J, Younis M I, Isa N A M and Abdullah R 2011 Design And
Implementation Of A T-Way Test Data Generation Strategy With Automated Execution Tool
Support Inf. Sci. (Ny). 181 1741–58
[16] H Y O and Kamal Z Z 2016 Development of interaction test suite generation strategy with input-
output mapping supports Sci. Res. Essays 6 3418–30
[17] Wang Z and He H 2013 Generating variable strength covering array for combinatorial software
testing with greedy strategy J. Softw. 8 3173–81
[18] Othman R R, Khamis N and Zamli K Z 2014 Variable Strength T-Way Test Suite Generator
With Constraints Support Malaysian J. Comput. Sci. 27 204–17
[19] Lin J, Luo C, Cai S, Su K, Hao D and Zhang L 2016 TCA: An Efficient Two-Mode Meta-
Heuristic Algorithm for Combinatorial Test Generation Jinkun Proc. - 2015 30th IEEE/ACM Int.
Conf. Autom. Softw. Eng. ASE 2015 494–505
[20] Ahmed B S, Abdulsamad T S and Potrus M Y 2015 Achievement of Minimized Combinatorial
Test Suite for Configuration-aware Software Functional Testing Using The Cuckoo Search
algorithm Inf. Softw. Technol. 66 13–29
[21] Yang X S and Deb S 2009 Cuckoo search via Lévy flights 2009 World Congr. Nat. Biol. Inspired
Comput. NABIC 2009 - Proc. 210–4
[22] Nasser A B, Sariera Y A, Alsewari A R A and Zamli K Z 2016 Assessing Optimization Based
Strategies for t-way Test Suite Generation: The Case for Flower-based Strategy Proc. - 5th IEEE
Int. Conf. Control Syst. Comput. Eng. ICCSCE 2015 150–5
[23] Zamli K Z, Alkazemi B Y and Kendall G 2016 A Tabu Search hyper-heuristic strategy for t-way
test suite generation Appl. Soft Comput. J. 44 57–74
[24] Ali M S A R, Othman R R, Yahya Z R, Ahmad M Z Z and Ramli N 2017 Implementation of
Artificial Bee Colony Algorithm for T-Way Testing 2016 3rd International Conference on
Electronic Design, ICED 2016 (Institute of Electrical and Electronics Engineers Inc.) pp 591–4
[25] Zamli K Z, Din F, Baharom S and Ahmed B S 2017 Fuzzy Adaptive Teaching Learning-Based
Optimization Strategy for The Problem of Generating Mixed Strength T-Way Test Suites Eng.
Appl. Artif. Intell. 59 35–50
[26] Dorigo M and Di Caro G 1999 Ant Colony Optimization: A New Meta-Heuristic Proc. 1999
Congr. Evol. Comput. CEC 1999 2 1470–7
[27] Dorigo M and Stutzle T 2010 Ant Colony Optimization: Overview and Recent Advances vol 146
(Springer International Publishing AG)
[28] Dorigo M 1997 Ant Colony System: A Cooperative Learning Approach to the Traveling
Salesman Problem IEEE Trans. Evol. Comput. 1 53

10