SCHEDULING AND INSPECTION PLANNING IN SOFTWARE DEVELOPMENT PROJECTS USING MULTI-OBJECTIVE HYPER-HEURISTIC EVOLUTIONARY ALGORITHM

International Journal of Software Engineering & Applications (IJSEA), Vol.4, No.3, May 2013
DOI : 10.5121/ijsea.2013.4304 45
SCHEDULING AND INSPECTION PLANNING IN
SOFTWARE DEVELOPMENT PROJECTS USING
MULTI-OBJECTIVE HYPER-HEURISTIC
EVOLUTIONARY ALGORITHM
A.Charan Kumari1
and K. Srinivas2
1
Department of Physics & Computer Science, Dayalbagh Educational Institute,
Dayalbagh, Agra, India
charankumari@yahoo.co.in
2
Department of Electrical Engineering, Dayalbagh Educational Institute, Dayalbagh,
Agra, India
ksri12@gmail.com
ABSTRACT
This paper presents a Multi-objective Hyper-heuristic Evolutionary Algorithm (MHypEA) for the solution
of Scheduling and Inspection Planning in Software Development Projects. Scheduling and Inspection
planning is a vital problem in software engineering whose main objective is to schedule the persons to
various activities in the software development process such as coding, inspection, testing and rework in
such a way that the quality of the software product is maximum and at the same time the project make span
and cost of the project are minimum. The problem becomes challenging when the size of the project is
huge. The MHypEA is an effective metaheuristic search technique for suggesting scheduling and inspection
planning. It incorporates twelve low-level heuristics which are based on different methods of selection,
crossover and mutation operations of Evolutionary Algorithms. The selection mechanism to select a low-
level heuristic is based on reinforcement learning with adaptive weights. The efficacy of the algorithm has
been studied on randomly generated test problem.
KEYWORDS
Scheduling and Inspection planning, software project development, Multi-objective optimization, Hyper-
heuristics, Evolutionary Algorithm
1. INTRODUCTION
The planning of software projects involves various intricacies such as the consideration of
resources, scheduling of members, dependencies among various activities, project deadlines and
many other restrictions. Besides these restrictions, the waiting times and other external
uncertainties also complicate the planning process. Due to the increase in the size and complexity
of the software projects, it is becoming a difficult task for the project managers to have a control
on the development costs and to maintain the deadlines of the projects. These two factors are in
turn dependent on the efficient scheduling of members to various activities involved in the
software development process such as coding, inspection, testing etc. Thus, the main goal of
scheduling and inspection planning process are to achieve high quality software product with
minimum development cost and project make span.

46
Search-based approaches to software project management were investigated by many researchers.
In 2001, Carl K. Chang and et al. [1] developed a new technique based on genetic algorithms
(GA) that automatically determines, using a programmable goal function, a near-optimal
allocation of resources and resulting schedule that satisfies a given task structure and resource
pool. In their method, they assumed that the estimated effort for each task is known a priori and
can be obtained from any known estimation method such as COCOMO. Based on the results of
these algorithms, the software manager will be able to assign tasks to staff in an optimal manner
and predict the corresponding future status of the project, including an extensive analysis on the
time and cost variations in the solution space. The results of the GA algorithm were evaluated
using exhaustive search for five test cases. In these tests the proposed GA showed strong
scalability and simplicity.
In 2005, Thomas Hanne and Stefan Nickel [2] modelled the problem of planning inspections and
other operations within a software development (SD) project with respect to the objectives quality
(no. of defects), project make span, and costs, as a multi-objective optimization problem. Their
considered model of SD processes comprises the phases coding, inspection, test, and rework and
includes the assignment of operations to persons and the generation of a project schedule. They
developed a multi-objective evolutionary algorithm and studied the results of its application to
sample problems.
E. Alba and J. F. Chicano [3], tackled the general Project Scheduling Problem with genetic
algorithms. In their approach, they combined the duration and cost objectives into a single fitness
function using weights in such a way that one can adjust these fitness weights to represent
particular real world project. They developed an automated tool based on genetic algorithms that
can be used to assign people to the project tasks in a nearly optimal way trying different
configurations concerning the relative importance of the cost and duration of the project. They
have performed an in depth analysis with an instance generator and solved 48 different project
scenarios and performed 100 independent runs for each test to get statistically meaningful
solutions. Their experiments concluded that the instances with more tasks are more difficult to
solve and their solutions are more expensive and the projects with a larger number of employees
are easier to tackle and can be driven to a successful end in a shorter time.
Leandro L. Minku et al. [4] presented novel theoretical insight into the performance of
Evolutionary Algorithms (EA) for the Project Scheduling Problem. Their theory inspired
improvements in the design of EAs, including normalisation of dedication values, a tailored
mutation operator, and fitness functions with a strong gradient towards feasible solutions.
According to their findings, Normalisation removes the problem of overwork and allows an EA to
focus on the solution quality and facilitates finding the right balance between dedication values
for different tasks and allows employees to adapt their workload whenever other tasks are started
or finished. Their empirical study concludes that normalisation is very effective in improving the
hit rate, the solution quality and making the EA more robust.
This paper presents a hyper-heuristic based multi-objective evolutionary algorithm [5] for the
solution of scheduling and inspection planning in the software development process, based on the
model suggested by Thomas Hanne and Stefan Nickel [2]. We implemented twelve low-level
heuristics based on various methods of selection, crossover and mutation operators in
evolutionary algorithms. The designed selection mechanism selects one of the low-level
heuristics based on reinforcement learning with adaptive weights. The efficacy of the algorithm
has been studied on randomly generated test problem [2]. Through our experimentation we found
that MHypEA is able to explore and exploit the search space thoroughly, to find high quality

47
solutions. And also the proposed algorithm is able to achieve better results in half the amount of
time expended by the MOEA reported in the literature.
The rest of the paper is organised as follows. Section 2 describes the scheduling and inspection
planning as multi-objective search problem. The basic concepts of hyper heuristics are presented
in section 3. The description of the proposed approach is provided in section 4. Section 5 presents
the Experiments and the results obtained are presented and analysed in section 6. Concluding
remarks are given in section 7.
2. A MULTI-OBJECTIVE SCHEDULING AND INSPECTION PLANNING
PROBLEM
This section briefly describes scheduling and inspection planning as a multi-objective search-
based problem as proposed by Thomas Hanne et al. and the detailed description can be found in
[2]. The basic activities and their sequence in the model are given in Figure 1.
The basic assumption of the model is that there are ‘n’ modules to develop, each with known size
and complexity. All the activities involved in the process are done by a team of ‘m’ developers.
The first activity in the model is coding. Each module is coded by one developer called its author.
After a module is coded, it is inspected by a team of developers known as inspection team. Based
on the outcome of inspection, the author reworks on the module. Thereafter, the module
undergoes testing done by a developer called tester and again the author reworks on the module
based on the findings of the testing process. All the activities are performed as per the sequence
shown in Figure1. As per the model, inspections are assumed to be done in a pre-empting way,
forcing the developers to interrupt their coding or testing process to inspect a module, as soon as
it is coded by its author.
As per the assumption each module is coded by only one developer called its author. Every
module is inspected by an inspection team of size 0 to m-1, with a restriction that an author of a
module cannot be its inspector. Similarly, each module is tested by a tester, different from its
author. For each module, priority values are assigned for coding and testing activities to
determine the temporal sequence for scheduling the tasks. The three objectives [2] considered in
the model are described below.
2.1. First objective – Quality of the product
The first objective is the quality of the product, measured in terms of total number of defects
produced and is calculated by
∑=
=
n
i
idtd
1 (1)
where
rdfdrdfdpdd ikikiKiKiki
2211
+−+−=
(2)
pdik denotes the produced defects during coding of an module i by an author k and is assumed to
be
coding rework testing reworkinspection
Figure 1 Sequence of activities in software development

48
cqsmddcplxsizepd kiiik /..= (3)
where size and cplx are the size and complexity of an module i and mdd denotes the minimum
defect density and cqs is the coding quality skill of the author k.
fd1
iK determines the found defects in a module i by the inspection team K and is given by
)).1(1.(1
∏ −−=
∈Kk
kiiK ddsitfpdfd
(4)
where itf represents inspection technique factor and dds represents the defect detection skill of the
inspector k.
It is assumed that though the rework of a module by its author removes the found defects by the
inspection team, it may introduce new defects in proportional to the found defects and is given by
fdrdfrd iKiK
11
.= (5)
where rdf represents rework defects factor.
fd2
ik denotes found defects in a module i, when it is tested by a tester k, and is given by
)1.( ..'2
edfd tttqsdfr
iik
ik−−=
(6)
where d′ represents the defects remaining in a module i after coding, inspection and rework, dfr
denotes defect find rate, tqs is the testing quality skill of the tester k and tt represents test time of a
module i and is determined as
sizecplxttt iiii ..=
(7)
where ti is test intensity.
Similarly, it is assumed that though the rework of a module by its author removes the found
defects by the tester, it may introduce new defects in proportional to the found defects and is
given by
fdrdfrd ikik
22 .=
(8)
2.2. Second objective – Project make span
In order to compute the project make span, a complete schedule for the software development
process is to be made. For each module, the specific time of each activity along with its waiting
times are to be calculated and the maximum time among all the modules determines the project
make span. The basic assumptions made in the model are that there are no specific dependencies
among the coding operations and all the inspections are carried out without any waiting times.
The coding time for a module is calculated as
)./(. cpsmcpcplxsizect kiii =
(9)
where mcp corresponds to the maximum coding productivity and cps corresponds to the coding
productivity skill of the author k.

49
The inspection time for a module i by the kth
inspector is calculated as
)./(. ipsmipcplxsizeit kiiik =
(10)
where mip corresponds to maximum inspection productivity and ips corresponds to inspection
productivity skill of the inspector k. The inspection time for a module is taken as the maximum
inspection time taken among the members of the inspection team.
The rework times are calculated as
)../(.11 cpsmcpadscplxfdrt kiii =
(11)
)../(.22 cpsmcpadscplxfdrt kiii =
(12)
where rt1
i and rt2
i represents rework times after inspection and testing respectively and ads
corresponds to average defect size.
The waiting time of the activities depends on the temporal sequence of the modules based on
coding and testing priorities along with the availability of the developers to carry out the specific
activity.
2.3. Third objective – cost
The project costs are assumed to be proportional to the effort which is measured as the total time
taken for each activity. Thus the cost is calculated as
)(. 21
rtttrtitctctc iiiik
i
i +++∑+∑=
(13)
where c represents unit cost of effort.
Thus, the Multi-objective scheduling and inspection planning problem is to schedule the
developers to various activities of different modules, in such a way that the number of defects,
project make span and cost are minimum.
3. HYPER-HEURISTICS
Hyper-heuristics are often defined as “heuristics to choose heuristics” [6]. A heuristic is
considered as a rule-of-thumb that reduces the search required to find a solution. Meta-heuristic
operates directly on the problem search space with the goal of finding optimal or near-optimal
solutions; whereas the hyper-heuristic operates on the heuristics search space which consists of all
the heuristics that can be used to solve a target problem. Thus, hyper-heuristics are search
algorithms that do not directly solve problems, but, instead, search the space of heuristics that can
then solve the problem. Therefore Hyper-heuristics are an approach that operates at a higher level
of abstraction than a metaheuristic. The framework of the proposed hyper-heuristic approach is
shown in Figure 2 [5].
The term Hyper-heuristics was coined by Cowling et al. and described it as “The hyper-heuristics
manage the choice of which lower-level heuristic method should be applied at any given time,
depending upon the characteristics of the heuristics and the region of the solution space currently
under exploration” [7]. So, they are broadly concerned with intelligently choosing a right
heuristic. The main objective of hyper-heuristics is to evolve more general systems that are able

50
to handle a wide range of problem domains. A general frame work of a hyper-heuristic is
presented in Algorithm 1 [6].
Algorithm 1 Hyper-heuristic algorithm
1: Start with a set H of heuristics, each of which is applicable to a problem state and transforms it
to a new problem state.
2: Let the initial problem state be S0.
3: If the problem state is Si then find the heuristic that is most suitable to transform the problem
state to Si+1.
4: If the problem is solved, stop. Otherwise go to step 3.
4. PROPOSED APPROACH
This section explains our proposed approach [5]. The designed format of the low-level heuristics
is EA/selection/crossover/mutation. The selection process involves identifying the parents for
generating the offspring. Two types of selection are proposed – rand and rand-to-best. In rand,
both the parents are selected randomly from the population, while in rand-to-best, one parent is
selected randomly from the population and the other parent is elite (the best one) selected from
the set of elites. Three types of crossover operators are identified to generate the offspring. The
first operator is uniform crossover; where in the offspring is generated by randomly selecting
each gene from either of the parents. The second operator is a hybrid crossover1 (hc1) that is
defined by hybridizing the single-point crossover with uniform crossover. The third operator is a
hybrid crossover2 (hc2) that is framed by the hybridization of two-point crossover with uniform
crossover. Two types of mutation are proposed - copy and exchange. In the first mutation
operator, two genes are selected randomly and the second gene is copied into the first one. In the
second mutation operator, two randomly selected genes exchange their positions. Based on the
above discussed selection, recombination and mutation operators, twelve low-level heuristics are
proposed for the hyper-heuristic and are shown in Table 1 [5].
CR ≤ 0.5 CR > 0.5
DOMAIN BARRIER
HYPER-HEURISTIC
LOW-LEVEL HEURISTICS
…..
Objective functions
h1
1
h2
1
1
h12
1
CR = U (0.0, 1.0)
hb = 1
he = 6
hb = 7
he = 12
hselect = roulette(hb, he)
Figure 2. Framework of the proposed Hyper-heuristic

51
The proposed hyper-heuristic selects a capable low-level heuristic in all iterations based on the
information about the efficacy of each low-level heuristic accrued during the preceding iterations.
This is executed through the principle of reinforcement learning [8]. The basic idea is to “reward”
improving low-level heuristics in all iterations of the search by increasing its weight gradually
and “punish” poorly performing ones by decreasing its weight. The weights of low-level
heuristics are changed adaptively during the search process and at any point they reflect the
effectiveness of low-level heuristics.
Table 1 Set of low-level heuristics used by the proposed hyper-heuristic
Group with copy mutation Group with exchange mutation
h1 : EA/rand/uniform/copy h7 : EA/rand/uniform/exchange
h2:EA/rand-to-best/uniform/copy h8 : EA/rand-to-best/uniform/ exchange
h3 : EA/rand/hc1/copy h9 : EA/rand/hc1/ exchange
h4 : EA/rand-to-best/hc1/copy h10 : EA/rand-to-best/hc1/ exchange
h5 : EA/rand/hc2/copy h11 : EA/rand/hc2/ exchange
h6 : EA/rand-to-best/hc2/copy h12 : EA/rand-to-best/hc2/ exchange
In the beginning all the low-level heuristics are assigned with equal weight. The weight of a
heuristic is changed as soon as a heuristic is called and its performance is evaluated. If the called
heuristic lead to an improvement in the values of the objective functions, its weight is increased,
otherwise it is decreased. All the weights are bounded from above and from below. Thus the
current values of the weights indicate the information about the past experience of using the
corresponding heuristics. The roulette-wheel approach is used to select a heuristic randomly with
the probability proportional to its weight [9].
The proposed hyper-heuristic works in two phases. The first phase selects the type of mutation to
be adopted (copy or exchange). This is done randomly with equal probability of both the groups
being selected. The values of hb and he denotes the subscripts of the beginning and ending of the
selected low-level hyper-heuristics. The second phase explicitly selects a low-level heuristic
within the selected group. This phase uses the reinforcement learning approach with adaptive
weights using roulette-wheel to select a particular model of EA. The framework of the proposed
hyper-heuristic is shown in Figure 2 [5]. Here, CR is a random number drawn from a uniform
distribution on the unit interval, to select an EA model either with copy or with exchange
mutation with equal probability.
Based on the selected low-level heuristic, an offspring population is generated and its fitness is
evaluated. Thereafter, the parent and offspring populations are combined together and a non
dominated sorting [10] is performed on the combined population to classify the solutions into
different Pareto fronts based on the goodness of the solutions. The population for the next
iteration is taken from the best fronts. The pseudo code of the MHypEA is given in Algorithm 2.

52
Algorithm 2 Multi-objective Hyper-heuristic Evolutionary Algorithm (MHypEA) [5]
1: Initialize parent population
2: Evaluate the fitness of parent population
3: While (not termination-condition) do
4: Select a low-level heuristic based on the selection mechanism
5: Apply the selected low-level heuristic on the parent population and obtain offspring
population
6: Evaluate the offspring population
7: Combine parent and offspring populations
8: Perform non dominated sorting on the combined population and select the individuals
from the best fronts for the next iteration
9: end while
5. EXPERIMENTS
This section describes methodology applied to compute the values of objective functions, test
problem considered along with the problem and algorithmic parameter settings.
5.1 Methodology
In this subsection we describe the initialization of decision variables, method adopted for
crossover and mutation operations and the procedure for the evaluation of objective functions.
5.1.1 Initialization of decision variables
The decision variables taken are (author, no_inspectors, inspector, tester, coding_priority,
testing_priority). Initially the developers are assigned randomly as authors to all the modules with
a condition that each module is coded exactly by only one author and an author may be assigned
to code more than one module. The number of inspectors for a module characterizes the
inspection team size of that module which is initially assigned randomly with a value in the range
0 to (m-1), where m signifies number of developers involved in the project; the upper limit of
(m-1) indicates that an author of a module cannot be its inspector and a lower limit of 0 indicates
that the module is not subject to inspection. In the current scenario, the maximum number of
inspectors for each module is taken as 3. The inspectors are randomly assigned for each module
as dictated by the inspection team size. Similarly, a developers are assigned as testers for testing
modules, with a constraint that an author of a module cannot be its tester. Further,
coding_priority and testing_priority for the modules are taken as decision variables and are
assigned values in the range [0, 1], to determine the temporal sequence for scheduling operations;
with an intention that a module with a higher priority value is to be scheduled first than a module
with a lower priority value.
5.1.2 Crossover operator
Parents are selected for crossover operation in two ways – rand and rand-to-best. Then for each
module in the offspring the author, inspectors, tester, priorities of coding and testing operations
are assigned with the values of a randomly chosen module from either of the parents, in
accordance with the three methods discussed in section 4. In this way the offspring are generated
from the parents leading to different permutations of scheduling.

53
5.1.3 Mutation
A simple strategy is adopted for mutation. In every offspring that is generated from the crossover
operation, two modules are selected randomly and the scheduling assignment of one is copied
into another in copy variant of mutation and the scheduling assignments are exchanged between
the selected modules in the case of exchange variant of mutation.
5.1.4 Evaluation of objective functions
The number of defects indicating the quality of the product and the cost objectives can be
calculated straightforwardly as per the formulae depicted in section 2. But the calculation of
project make span is a complicated one, as one needs to frame the complete schedule of the
software development process. The approach used in this paper for the computation of project
make span is as follows: as a first step, the time for each activity, (depicted in the Figure 1) for
each module is calculated. In the next step, the waiting times are calculated before each activity
starts for each module, based on observing the given assignment of persons to modules &
activities and scheduling the modules for one person according to the priorities, coding_priority
for coding and rework, testing_priority for tests. Inspections are assumed to be performed as soon
as the coding finishes, without any waiting time. Since inspection activity has a higher priority
than other activities, it is assumed that they interrupt the coding activity of an inspector. All the
activities are scheduled by considering the restrictions on their precedence. Finally the make span
of each module is computed based on the actual activity times and their corresponding waiting
times. The maximum make span among the modules is considered as the project make span.
5.2 Test Problem
In order to evaluate the efficiency of the proposed MHypEA, test problem and the problem
parameters are taken from the literature as recommended in [2]. The following technical
parameters are used:
• number of modules: n = 100
• number of developers: m = 20
• maximum coding productivity: mcp = 25 [loc/h]
• minimum defect density: mdd = 0.02 [defects/loc]
• maximum inspection productivity: mip = 175 [loc/h]
• inspection technique factor: itf = 0.45
• test intensity: ti = 0.05
• defect find rate: dfr = 0.1
• rework defects factor rdf = 0.1
• average defect size: ads = 8 [loc]
• unit costs : c = 150[EUR/h]
For the all skill attributes, a normal distribution with an expected value 0.5 and a variance of 0.1
is assumed (but ensuring that the values are in [0, 1]); with an assumption that the person on the
average reach out to 50% of the optimum skill values. For the module size, a lognormal
distribution with expected value 300 [loc] and variance 120 is applied. For the module
complexity, a normal distribution with expected value 1 and variance 0.1 is assumed.
The population size is taken as 30 and the test problem were run for a maximum of 500 iterations.
The algorithm has been implemented in MATLAB 7.6.0, on an Intel® Core™ 2 Duo CPU T6600
@2.20 GHz processor, 3 GB RAM and Windows 7 platform.

54
6. RESULTS
The results obtained by MHypEA on the above described test problem is presented in this section.
Figures 3-5 represents the box plots of some generations visualizing the distribution of the three
objective functions.
The box plots of figures 3 and 4 shows the most considerable improvement in the best values for
the defects objective within the first 150 generations and costs objective within the first 250
generations of MHypEA. With respect to the duration objective there is a negligible improvement
in the best values found. The conflicting nature of the objectives is evident from the three box
plots.
50 100 150 200 250 300 350 400 450 500
175
180
185
190
195
200
205
210
215
220
Defects
Geneartion
Figure 3. Defects of solutions based on generation number
50 100 150 200 250 300 350 400 450 500
6.15
6.2
6.25
6.3
6.35
6.4
6.45
6.5
6.55
6.6
x 10
5
Costs
Generation
Figure 4. Costs of solutions based on generation number

55
50 100 150 200 250 300 350 400 450 500
280
300
320
340
360
380
400
420
440
Generation
Duration
Figure 5. Duration of solutions based on generation number
6.15 6.2 6.25 6.3 6.35 6.4 6.45 6.5 6.55
x 10
5
175
180
185
190
195
200
205
210
215
costs
defects
gen. 100
gen. 200
gen. 300
gen. 400
gen. 500
Figure 6. Costs and Defects in solutions of several generations
6.15 6.2 6.25 6.3 6.35 6.4 6.45 6.5 6.55
x 10
5
260
280
300
320
340
360
380
400
420
440
co sts
Duration
g en . 100
g en . 200
g en . 300
g en . 400
g en . 500
Figure 7. Costs and Duration in solutions of several generations

56
260 280 300 320 340 360 380 400 420 440
175
180
185
190
195
200
205
210
215
Duration
defects
gen. 100
gen. 200
gen. 300
gen. 400
gen. 500
Figure 8. Duration and Defects in solutions of several generations
Figures 6-8 represents the solutions obtained by MHypEA for some selected generations for each
combination of two objectives. It is apparent from the three figures 6-8 that there is a continuous
progress towards the minimization of objectives and the diversity of the solutions. For example,
considering the population of generation 500 there is a significant improvement in the number of
defects and costs, in comparison to the population of generation 100. As the objectives are highly
conflicting, in some instances one objective is decreasing only by increasing the other objective.
The comparison of the results of MHypEA with the Multi-objective Evolutionary Algorithm
(MOEA) reported in the literature [2] shows that MHypEA is able to achieve better values for all
the three objective functions in half the number of generations. This is perceptible from the
corresponding box plots and 2-dimensional plots of MOEA [2] and MHypEA.
7. CONCLUSION
This paper presents a Multi-objective Hyper-heuristic Evolutionary Algorithm for the solution of
scheduling and inspection planning in software development projects. Scheduling and inspection
planning is an important problem in the software development process, as it directly influences
the quality of the end product. The scheduling of personnel is considered for coding, inspection,
testing and rework phases of the development process. Three highly conflicting objectives of
number of defects found (indicating the quality of the product), costs and duration of the project
are evaluated. A hyper-heuristic based multi-objective evolutionary algorithm is proposed for the
purpose and the results are assessed. The obtained results show that the proposed algorithm is
able to improve the solutions significantly with better diversity in the solutions. The comparison
of the results with MOEA shows that MHypEA is able to achieve high quality solutions in half of
the number of iterations.
ACKNOWLEDGEMENTS
The authors are extremely grateful to the Revered Prof. P.S. Satsangi, Chairman, Advisory
Committee on Education, Dayalbagh for continued guidance and support.

57
REFERENCES
[1] C. K. Chang, M. J. Christensen, T. Zhang, (2001) “Genetic algorithms for project management”,
Annals of Software Engineering, Vol. 11, pp107-139.
[2] T. Hanne & S. Nickel, (2005) “A multiobjective evolutionary algorithm for scheduling and
inspection planning in software development project”, European Journal of Operational Research,
Vol. 167, pp 663-678.
[3] E. Alba & J. F. Chicano, (2007) “Software project management with GAs”, Information Science,
Vol. 177, pp 2380-2401.
[4] Leandro L. Minku, Dirk Sudholt, Xin Yao, (2012) “Software Evolutionary Algorithms for the Project
Scheduling Problem : Runtime Analysis and Improved Design”, Proceedings of the fourteenth
international conference on Genetic and evolutionary computation conference, pp 1221-1228.
[5] Charan Kumari, A. & Srinivas, K., (2013) “ Software Module Clustering using a Fast Multi-objective
Hyper-heuristic Evolutionary Algorithm”, International Journal of Applied Information Systems,
Vol. 5, pp 12-18.
[6] Burke, E., Kendall, G., Newall, J., Hart, E., Ross, P., and Schulenburg, S. (2003) “Hyper-heuristics:
an emerging direction in modern search technology”, Handbook of metaheuristics, pp. 457–474,
Kluwer Academic Publishers.
[7] Cowling, P.I., Kendall, G., Soubeiga, E. (2001) “Hyperheuristic Approach to Scheduling a Sales
Summit”, Proceedings of the Third International Conference of Practice And Theory of Automated
Timetabling, Vol. 2079, pp 176-190.
[8] Kaelbling, L.P., Littman, M.L., Moore, A.W. (1996) “Reinforcement learning: a survey”, Journal of
Artificial Intelligence Research 4, 237–285.
[9] Nareyek,A. (2003) “Choosing search heuristics by non-stationary reinforcement learning”, In
Metaheuristics: Computer decision-making, pp. 523–544. Kluwer Academic Publishers.
[10] K., A. Pratap, S. Agarwal, and T. Meyarivan. (2002) “A Fast and Elitist Multiobjective Genetic
Algorithm: NSGA-II”, IEEE Transactions on Evolutionary Computation, Vol. 6, pp 182-197.
Authors
A. Charan Kumari received the Master of Computer Applications degree from Andhra
University, India in 1996 and stood first in the district, and M. Phil degree in Computer
Science from Bharathidasan University, Tiruchirappalli, India. She has an excellent
teaching experience of 15 years in various esteemed institutions and received various
accolades including the best teacher award from the management of D. L. Reddy College,
Andhra Pradesh. She is working towards the Ph.D. degree in Computer science at
Dayalbagh Educational Institute, Agra, India under collaboration with IIT Delhi, India
under their MoU. Her current research interests include Search-based Software engineering, Evolutionary
computation and soft computing techniques. She has published papers in international conferences and
journals. She is a life member of Systems Society of India (SSI).
K. Srinivas received B.E. in Computer Science & Technology, M.Tech. in Engineering
Systems and Ph.D. in Evolutionary Algorithms. He is currently working as Assistant
Professor in Electrical Engineering Department, Faculty of Engineering, Dayalbagh
Educational Institute, Agra, India. His research interests include Soft Computing,
Optimization using Metaheuristic Search Techniques, Search-based Software
Engineering, Mobile Telecommunication Networks, Systems Engineering, and E-
learning Technologies. He received Director’s medal for securing highest marks in
M.Tech. Programme at Dayalbagh Educational Institute in 1999. He is an active
researcher, guiding research, published papers in journals of national and international repute, and is also
involved in R&D projects. He is a member of IEEE and a life member of Systems Society of India (SSI).
He is also the Publication Manager of Literary Paritantra (Systems)- An International Journal on Literature
and Theory

SCHEDULING AND INSPECTION PLANNING IN SOFTWARE DEVELOPMENT PROJECTS USING MULTI-OBJECTIVE HYPER-HEURISTIC EVOLUTIONARY ALGORITHM

More Related Content

What's hot (19)

Viewers also liked (20)

Similar to SCHEDULING AND INSPECTION PLANNING IN SOFTWARE DEVELOPMENT PROJECTS USING MULTI-OBJECTIVE HYPER-HEURISTIC EVOLUTIONARY ALGORITHM (20)

Recently uploaded (20)

SCHEDULING AND INSPECTION PLANNING IN SOFTWARE DEVELOPMENT PROJECTS USING MULTI-OBJECTIVE HYPER-HEURISTIC EVOLUTIONARY ALGORITHM