International Journal of Computational Intelligence Systems, Vol.

7, Supplement 1 (2014), 121-136

Applicability of artificial bee colony

algorithm for nurse scheduling problems

Kadir Buyukozkan
Industrial Engineering Department, Karadeniz Technical University,
Kanuni Campus, 61080, Trabzon, Turkey

Ahmet Sarucan
Industrial Engineering Department, Selçuk University,
Alaaddin Keykubat Campus, 42000, Selçuklu/Konya, Turkey
E-mail: [email protected]
www.selcuk.edu.tr

Received 15 December 2012

Accepted 1 July 2013

Abstract
This paper describes the first Artificial Bee Colony (ABC) Algorithm approach applied to nurse scheduling
evaluated under different working environments. For this purpose, the model has been applied on a real hospital
where data taken from different departments of the hospital were used and the schedules from the model were
compared with the existing schedules. The results obtained indicated that the proposed model exhibits success in
solving the nurse scheduling problems in hospitals.
Keywords: Artificial bee colony algorithm, nurse listing, nurse scheduling, task assignment

physical conditions must be exact and planned correctly

1. Introduction to improve service quality in health sector.
Personnel scheduling, aims at satisfying appropriate One of the most important conditions in any well
labor needs of enterprises for their sustainable business. functioning health enterprise is employees with enough
The procedure has been subjected in literature as knowledge and ability. In this case, personnel
following different names; labor scheduling, worker scheduling studies in health sector becomes more
scheduling, shift scheduling, task assignment, crew important than in other service sector and is being
scheduling and crew planning subject in academic studies more than others.
Health sector is an important part of service industry. Controlling all solution points for the solution of the
Service presented in this sector contains operations in NP-Hard problems with wide solution spaces like nurse
patient’s medical examination processes. Doing the scheduling problems is not feasible in an acceptable
work incorrectly and not at its time, in the other words time, even with today’s technology. Heuristic methods
not to present the service as needed, may cause control some of the points in solution space with respect
irreversible worse situations. Because of that, service to several rules and can find an acceptable solution for
quality in health sector bears special importance. All the problem in a short time period. In last terms,
numerous studies with heuristic methods prove its

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
121
K. Buyukozkan, A. Sarucan

success and reliability on this subject. Besides these Colony (ABC) algorithm has been developed. In the
features, another important advantage of heuristic second section of the paper, specifications of NSP have
methods is that, it is not hard to adapt and model it for been explained in detail. In third section, artificial bee
different problem types. colony algorithm is introduced and description on how
Meta-heuristic methods used in literature for the nurse to tailor it to the nurse scheduling problems is given. In
scheduling problem are; Genetic Algorithm, the fourth section experimental results have been
Simulation, Tabu Search, Shifting Neighbor Search and shown. In the last part, we conclude with evaluation of
hybrid algorithms that combined different solving results and further research.
models.
Valouxis and Housos1, Burke et al.2, Burke et al.3, Li et 2. Nurse Scheduling Problems (NSP)
al.4, suggest hybrid algorithms for nurse scheduling. With the knowledge that nurses work at so many
Ref. 1 used Integer Linear Programming with Local different branches in health enterprises. In this study the
Search model and Tabu Search algorithm together. nurses that work in polyclinic services are taken into
Ref.2 hybridized Tabu search heuristic with Genetic consideration. Working conditions, mission and
algorithm. Ref. 3 used Heuristic Ordering and Shifting responsibilities of nurses have been established with
Neighbor Search together. Ref. 4, used goal related laws and status. Generally, nurses work 40
programming and meta-heuristic search approaches hours per week. Because of the obligation of working
together. 7/24, nurses become obliged to complete their weekly
Aickelina and Dowsland5, Bellanti et al.6, Burke et al.7, works within weekdays, during weekends, mornings,
Gutjahr and Rauner8, Beddoe et al.9, Çivril10, Bilgin et evenings and even at nights. This working regulation
al.11, Burke et al.12 developed different models for contains differences that hospital can identify itself. In
solving real-world nurse scheduling problems and made this case, some enterprises determine 8 hour - 3 shift
implementation studies. Aickelina and Dowsland5 and working regulation while others apply 8-hour and 16-
Çivril10 developed Genetic Algorithm models. Bellanti hour 2 shift system. As similar to that, nurses in one
et al.6 developed a Greedy Algorithm based local hospital can work both in day and night shifts while in
Neighbor Search approach. Burke et al.7 developed another hospital they can work continuously in day or
Tabu Search algorithm. Gutjahr and Rauner8 developed continuously at night.
Ant Koloni Algorithm for nurse scheduling for the first Although it is needed to get the work done, flexibility
time. Beddoe et al.9 used CABAROST (Case Based of working conditions in a wrong planned place leads to
Rostering) system which is a new model for nurse inevitable psychological and physical difficulties for the
scheduling. Bilgin et al.11 developed Local Neighbor workers. For example, because of the personnel
Search model. Burke et al.12 suggested a hybrid model insufficiency and other reasons, the remaining nurses
combined Integer Programming with Shifting Neighbor can work overtime. In this case a 16-hour night worker
Search algorithm. starts to work again without having sufficient rest.
Dias et al.12, Chiaramonte and Chiaramonte14, Similar to that, although she/he doesn’t want to, a nurse
Maenhout and Vanhoucke15 improved new heuristic becomes obliged to work continuously as a night sentry.
models for nurse scheduling problem. Cheng et al.16, Consistency of different working structures like intern
Parr and Thompson17, used simulated annealing method nurses and responsible nurses in the nurse crew is
for nurse scheduling. Maenhout and Vanhoucke18 and getting more difficult to planning studies. For this
Brucker et al.19, set new data sets to compare suggested reason, nurse scheduling has been examined in a
different models. Maenhout and Vanhoucke20, Tsai and number of studies in literature. Cheang et al.23
Li21 proposed various genetic algorithms in their studies mentioned that constraints of nurse scheduling came
for nurse scheduling. Oughalime et al.22 developed a from nurse contracts, statutory restrictions, nurse
solution model with Tabu Search heuristic. Cheang et demands and hospital specialties. They explained that
al.23, Causmaecker and Berghe24 made a detailed some of these constraints cannot be ignored but some of
literature search examining nurse scheduling studies. them can be flexible.
In this study, work on Nurses Scheduling Problem It is obligated to generate a good working schedule to
(NSP) was conducted. For NSP, an Artificial Bee minimize negative effects of working conditions on

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
122
 Bee Colony for Nurse Scheduling

personnel. In lots of health enterprises of our country, it will be understood that i nurse is not working at j day
nurse scheduling studies has been made by responsible k. shift, when it is equal to 1 she/he will be working.
nurses with hand. Insufficient personnel number, Nurse- Task Structure model is similar to Nurse-Day
specific working regulation and demands of the nurses model. Decision variable for every single shift is
make the scheduling problem more complicated. In this defined as vi,s .In this definition 1iN shows nurse
situation, a schedule that both satisfies workers number and 1sZ shows task number in scheduling
demands and needs of the enterprises is hard to prepare period. One difference between these two models is that
and it becomes a time consuming task. Addition to that, nurse-task shift definition doesn’t have to fit day so it
although it doesn’t seem as a problem formerly, can start in any time of day and finish in other day.
reflection of relations in people to scheduling is a Nurse-Shift Model Structure has a different form from
possible situation. two the models described above. Decision variable is
The fact that scheduling study has a dynamic nature defined as vi,p in this model. In this definition while
makes the situation more difficult. Nurse schedules are 1iN shows nurse number, 1 pM shows shift model
prepared every week or in general, monthly. Also number. In addition to that, every single shift model
demands of the employees make the schedule to have to shows the works that will make by a nurse through a
be rearranged occasionally. This repetitive workload planning term. In this type of problems if i. nurse
makes the scheduling more time-consuming and an works in p. shift type, decision variable gets value 1;
exhausting situation. else it gets 0 value.
The difficulties mentioned above bottlenecks in
preparing a qualified nurse scheduling, so this situation 2.1. Constraints used
hinders service quality. Usage of computer based When the literature is investigated, the major problem
models in nurse scheduling problems is very important in nurse scheduling sector is found to exist when
to create schedules that make possible both qualified comparison of proposed model with the previous works
service and time advantage. There are several models is made. In the study by Maenhout and Vanhoucke18, it
for the nurse scheduling suggested in the literature. was found that problem definition and modeling in their
Cheang et al.23, explain that a general nurse scheduling study are completely different from the ones found in
problems can be defined as one of the Nurse-day, literature, and that it is difficult to make comparisons of
nurse- task (nurse - time intervals) or nurse-shift type the proposed periods due to the fact that every hospital
models. has its own characteristic features and procedures hence
Nurse-Day structure is defined as a two dimensioned existence of differences. Apart from that, it was found
task list. According to that decision variable is that the exact and flexible constraints investigated are
determined for a nurse as vi,j. In this definition 1i N of a diverse types and that it is possible to take into
is nurse number, 1 j  P is day number in planning account the constraints of different functions. The study
term. Vi,j variable mentions that will be work at that day also found that the above explained issues clarify the
or not. At the same time shift number can be change, reason as to why the nurse scheduling studies have been
shifts can be work in a day created generally in dealt with differently in literature and that wide solution
literature as morning(A), noon(P) and night(N). Not- procedures have been proposed.
working situation generally get values as following: In the literature study by Ref. 23, the researchers
using permission (O), not coming but finding someone grouped the exact and flexible constrains, often used in
else instead (CO-compensation of), public holiday (PH) nurse scheduling under certain titles and made the
and values etc. In some studies, problem’s complexity following comments;
is reduced a bit by defining one type value for not- 1. Nurse working hours (minimum or maximum
working situation. For 0-1 models vi,j decision variable working time)
can be specified as vi,j,k. In this case 1kZ value will 2. Successive working within same shift (at least, at
show shift type number. So when vi,j,k value equals to 0 most or at specified amount)

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
123
K. Buyukozkan, A. Sarucan

3. Number of successive working days or shifts (at to this, if the difference of extra working hours of two
least, at most or at specified amount) nurses is 1, then penalty value is calculated by
4. Nurse education level and specialization area multiplying penalty constant by 1 whereas the penalty
5. Nurse demands and priorities constant is multiplied by 4 if the difference is 2.
6. Nurses’ vacation days (minimum, maximum, or
successive resting days) 2.3. Schedule feasibility
7. Spare times in between shifts ( minimum value) In order for the established schedule to be suitably
8. Shift type promotions (maximum shift type, amount accepted, it is necessary that the schedule covers all
needed for every shift en) needs of the department. An example of the nurse
9. Vacation days demands is shown in Table 1. As seen from the table,
10. Working on weekends (e.g. no working over the the demand inputs have been set on the basis of three
whole weekend) shifts as 8*16, 16*24 and 24*8. In order to meet these
11. Constrains related to nurse group or type (like some demands, the nurses can work on one of these shifts:
nurses can never work at the same time) 8*16, 8*24, 16*24, 16*8 and 24*8. If a nurse works at
12. Shift model constraints shift 16*8, for instance, she will have covered demands
13. Old records (promotion status on a previous from both shift 16*24 and shift 24*8.
schedule) In the conducted study, the model formed, first
14. Short and long term demands out of planning phase distributes all the demands to the nurses randomly, then
(e.g. a daily assignment to a shift) it runs the Artificial Bee Colony Algorithm to improve
15. Valid constraints throughout the shift (like no its solution quality based on exact and flexible
working on two shifts at the same time) constraints. When the algorithm tries to improve the
16. The need of a nurse for any shift (minimum, solution quality, it designates status by checking the
maximum or specified number) statuses that are contrary to the suitable solution status
In this study, the grouping specified in the study of and in this way; new schedules are formed without
Cheang et al23 has been taken consideration based on 1, impairing the suitability of the existing schedule.
5, 6, 7, 10, 12, 13, 15, 16 constraint types.
3. Artificial Bee Colony (ABC) Algorithm
2.2. Evaluation function
ABC algorithm is one of heuristic methods that started
In intuitional methods applied for nurse scheduling improving in the latest years. As in other colony
problem, the least penalized schedule is chosen as a algorithms, basic logic of the method is to get benefit of
solution by punishing an unwanted event. For this property in solving the encountering problems about
reason, it is necessary to specify penalty points for food finding of creatures living as colony in a minimum
situations deemed as constraints. Based on this, in case time interval with a good coordination although being
of unwanted event, it is then punished by multiplying it in complicated and crowded structure. ABC algorithm
with a specified constant. is a model trying to imitate natural properties of bees
When determining the punishment constants, the study used when they look for food.
conducted by Ref. 3 was taken into account. For
unwanted situations known as hard constrains, constant 3.1. Working logic of artificial bee colony
values ranging between 500 and 1000 were selected algorithms
whereas situations that could be allowed to occur also
Basic logic of the ABC algorithm is to prove feasible
termed as soft constrain, the constant values employed
solution in a short time period and prevent from
were varying between 50 and 150.
hanging local optima making search for more than one
Another unwanted situation that needs to be taken into
solution at the same time with a good neighbor search.
account is the constraints that appear more than once
With this thought every solution point is defined as a
which is worse than occurring once. For this reason,
bee. After that with having basis of strategy about
constants for these situations increase exponentially
transporting most productive and easy food to the hive
rather than linearly. Equal distributed constraint of extra
by bees in their natural life, a trial in finding the most
working hours can be given as an example. According
profitable point is made.

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
124
 Bee Colony for Nurse Scheduling

Table 1. Sample of a weekly nurse demands.

Monday Tuesday Wednesday Thursday Friday Saturday Sunday

Shifts 8-16(Day) 4 4 4 4 4 2 2
16-24(Night) 3 3 3 3 3 2 2
24-8(Night) 2 2 2 2 2 2 2

Examining widely interaction process between nature increased. When the sources were changed, the
and bees will be beneficial to fully understand the logic direction of the bees also changed. Movement of bees
of algorithm. Baykasoğlu et al.25 examined this process based on the changing locations of the sources is shown
in their study in two components: food sources and on Figure 1.
foragers as follows. (Ref. 25);
 Food Sources; Changes according to different
parameters like resource values; distance to the hive,
quality of nectar and ability of extrication level of
nectar.
 Forager Bees;
 Unemployed foragers; bees without any
knowledge on food sources. These bees can
search new sources of food as scout bees or
reach a food source by following other bees
 Employed foragers; when the unemployed bee Fig. 1. Selection in between nectar sources done by the colony
finds and exploits the food source, it will rise (Ref. 26)
to be an employed forager. These bees return
to the hive with a little nectar and empty it to The reason for superiority of ABC is the parallelism it
the hive. After that they inform the rest of the uses during the decision phase. This is due to the fact
bees in the hive about the amount of the nectar. that many bees move at the same direction at the same
 Experienced foragers; these bees direct the time while communicating with each other in order to
colony by using historical information about solve the problem at hand in the most effective way
the places and quality of food resources. possible.
An experiment to observe the way the colony makes When the existing studies in literature are investigated,
selection among nectar sources was conducted by Ref. it is found that apart from minor changes and some
26. Two nectar sources were placed close to the colony, increments, the ABC based models formed possess the
one at 400 meters south and the other at 400 meters below mentioned trends in general. (Ref. 25);
north of the hive. 12 bees were trained to head towards
the north while 15 bees were trained to move to the (i)Define Parameters
south side. The sources were formed at different (ii)Establish initial solutions
qualities where the south source was made better than (iii)Calculate suitability value of each bee
the one at north. At noon, the sources were exchanged, (iv) Determine scout bees and form follower bees
where the south source was placed on the north and under every scout bee
vice versa. In the other morning, their places were (v) Compare every scout bee with its follower bees
reversed again and at noon they were changed again. (vi) Replace the follower bee that is better than scout
The experimental observation indicated that the number bee with the scout bee
of bees at the low quality source remained low while (vii) Repeat steps 3-6 until the required iteration
the number of bees moving to the quality source number is reached

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
125
K. Buyukozkan, A. Sarucan

3.2. Adaptation of ABC algorithm for NSP and the number of exits belonging to that situation. “Eq.
Using the parameters described below, ABC algorithm (1)” and “Eq. (2)” respectively, show bee punishment
can be modeled for NSP (Ref. 27); and nurse punishment.
B : Number of initial bees, PBPb   in  1 Pi {b=1,2,…,B} (1)
L : Number of scout bees,
T : Number of follower,
Pi   Z
z  1C z xCC z {i=1,2,…,n} (2)
n : Number of nurses,
m : Number of days,
3.2.3. Defining scout bees;
o : Number of shifts,
Z : Number of situation needed to be punished. Bees formed as initial solution are arranged in
PBb : Initial bees in population {b=1,2,…,B} ascending order based on calculated punishment values.
SBl : Scout bees in population. {l=1,2,…,L} Then, the first L bees are termed as scout
FBlt : Follower bee of l. leader bee. {t=1,2,…,T} bees.(SB1,SB2,…,SBL=PB1,PB2,…PBL {PBP1 < PBP2 <
PBPb : Punishment points of initial bees. … < PBPB } , SBP1, SBP2, …,SBPT = PBP1,
{b=1,2,…,B} PBP2,…,PBPT {PBP1<PBP2<…<PBPB})
SBPl : Punishment points of scout bees.
{l=1,2,…,L} 3.2.4. Neighbor search to define follower bees of
FBPlt : Punishment points of l leader bee’s follower scout bees;
{t=1,2,…,T} Follower bees are first grouped together with the scout
Ni : Nurses. {i=1,2,…,n} bees they follow, “Eq. (3)”, then by applying the
Dj : Days. {j=1,2,…,m} neighbor finding function “Eq. (4)”, the scout bees are
Sk : Shifts. {k=1,2,…,o} brought to neighbor status.
Pi : Punishment points of nurses. {i=1,2,…,n}
Cz : Punishment coefficient of situations FBlt  SBl t,l (3)
{z=1,2,…,Z}
FBlt  FNS ( FBlt ) t,l (4)
CCz : Counts of appeared situations needed to be
punished,
xijk : Reassignment situation of nurses. 3.2.5. Calculating suitability values of follower
bees;
{i=1,2,…,n},
FNS() : Neighbor searching function. The suitability value of every bee is calculated from the
Processing steps of the developed algorithm by using punishment value of that bee. The method used in
these parameters are as shown below; calculating bees’ punishment, this time is used for the
followers. (“Eq. (5)”)
3.2.1. Forming the bees randomly for initial
solution; FBPlt   in  1 Pi t, l (5)
Every bee indicates PBh=[Ni]n series formed by n
nurses. And Ni=[Xijk]nX(o*m) series has been defined for 3.2.6. Comparing the scout and follower bees;
every nurse. The index Xijk shows i. nurse working (or In this step each scout bee is compared with its follower
not working) on j. day and at k. shift. bees and if a follower bee is better than the scout bee,
then ( FBP1t  SBPl {t=1,2,…,T} l ) the follower bee
3.2.2. Calculating every bee’s suitability value;
is defined as the scout bee.
The suitability value for each bee is obtained from The algorithm is prolonged in this way until a required
punishment values of that particular bee. And the bee’s condition is reached. When the stopping criterion is
punishment value is calculated from the total reached the best scout bee is accepted as the result.
punishment values of the nurses forming that bee. The
nurses’ punishment values are calculated as products of
punishment coefficient of a given unwanted situation

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
126
 Bee Colony for Nurse Scheduling

3.3. Literature studies on artificial bee colony 4. Case Study

algorithm
The developed ABC algorithm was applied on five
Studies on ABC have recently increased, being used for different departments of Karadeniz Technical
the purpose of obtaining suitable solutions for many University Research Hospital under different conditions
problems in various sectors. In their study, Nakrani and and its performance was then evaluated.
Tovey28 developed a new model by making use of Arrangement of the nurses working schedule on the
similarities in problems encountered by bees when they shifts at different departments was made by the
collect nectar and those experienced by internet responsible nurses on duty at the hospital. In this way, it
presenters. They compared the effectiveness of the new is possible to achieve a more flexible working structure.
model with the Greedy algorithm. Yang29, investigated This flexibility provides a more efficient use of the man
the use of ABC function optimization in solving power. Every nurse in each hospital department is
problems and its comparison with Genetic Algorithm. appointed by the hospital management and only works
Baykasoğlu et al.25 tested the ABC to hawk problems on day shifts during week days. After the schedules
and presented its successfulness especially for small developed by responsible nurses are approved by the
and medium sized problems. Dehuri et al.30 used ABC hospital management, they are put into effects.
for the multiple campaigns assignment problem. They Responsible nurses prepare schedules for their own
compared the proposed model with the Random and departments with hand, print them out on computers
Independent methods. In the Ref. 31, ABC was and transfer them for service works. When the
developed for solving Assignment Problems. In the schedules are drafted, efforts are made to make sure
study, the effects of shifting, changing, double shifting that nurses’ demands and work leaves are taken into
and subtraction chain neighborhood structures on the account. Besides this, compulsory and flexible
performance of ABC were studied. Other researchers constraints are considered too.
are Lara et al.32 whose study concentrated on the use of The number of nurses for selected departments varied
ABC on “School Time Scheduling”. Davidovic et al.33 between 6 and 13. These numbers are specified by the
tried to find a solution to the problem of establishing an hospital management based on the work density in
independent static schedule by making use of ABC. respective sections.
Özbakır et al.34 studied the performance of ABC’s Generally; the nurses work on 8-16, 16-24, 24-8, 8-24
Advanced Assignment Problems where; beside small or 16-8 shifts. Beside this, when the details given in
and medium sized problems, large sized problems were Table 3 are studied, it is found that two of the
also measured and comparison with the existing studies departments where the applications were made seem to
in the literature was made. The study by Kashani et want to work on only 8-16 and 16-8 shifts and three of
al.35 proposed the use of ABC for the assignment them are on 8-16, 16-24, 8-24 and 16-8 shifts. Based on
scheduling problem in the distribution system and they the tenth exact constraint given in Table 4, the model
showed that the model is better than the Genetic developed tries to prevent the departments from
Algorithm based models. working outside their respective assigned shifts by
Detailed literature studies on ABC models can also be punishing the shifts appointments. In this case, if a
found from the studies of Kaur and Goyal36, nurse in the physical therapy department is assigned the
Teodorovic37, Suri and Snehlata38. The success of ABC 24-8 shift or if a nurse in the schedule of pediatric
in achieving solutions to optimization problems is department is assigned to one of the 16-24, 8-24 or 24-8
proved by the literature studies cited on the topic, so shifts, then the nurse is punished.
far. However; when the literature is investigated, it is Nurse shift schedules in the existing hospitals are
found that there is no substantial study conducted on drafted weekly, by the nurse on duty in a particular
ABC with the nurse scheduling problems in mind. department. Similarly; effort is generally made to
equally, distribute the nurses’ night shifts, overtimes
and week-end works within a four-week period. The

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
127
K. Buyukozkan, A. Sarucan

developed algorithm drafts the schedule compatible to with these shift types, the number ten exact constraint is
this system and within the one month period, it also used.
takes previous weeks’ schedules into consideration.
In order to determine the constraints to be dealt with 4.1. Solution of the problem with ABC algorithm
and how will they affect the problem in this study, the First of all, a random initial solution is established
work Ref. 3 was investigated and responsible nurses based on the number of the bees in the program. When
working at the hospital were also contacted to have the initial solution is established, user demands and the
their views on the matter. The constraints used in the number of existing nurses are taken into account, and
model, their weight values and evaluation of their then all the demands are randomly assigned to the
punishments are given in Table 2. nurses. The program establishes initial solution by
Application trial of the system was carried out in executing the following algorithm steps randomly up to
different departments and under five different working the number of bees.
conditions common to all the departments; where the
performance of the proposed algorithm was evaluated. (i)Turn up to the initial number of bees
Characteristic features and special limitations of the (ii)Turn until all the demands are met
departments where the application was made are given (iii)Choose one day at random
in Table 3. Information showing the day and number of (iv) Choose a shift at random
nurses working at which shifts is found on Table 4. (v) Choose a nurse at random
When the Table 3 is studied, it is seen that, for every (vi) If the demand of a chosen shift on that day is not
department the first nurse is the responsible nurse for met and if the chosen nurse has not been assigned
that department. The responsible nurses have work that shift before, then assign the chosen demand to
schedules that only show that they can work in day the chosen nurse, otherwise don’t assign.
times on weekdays. That’s why these nurses are (vii) If the criterion to stop is achieved, then stop, save
assigned works on day shifts within weekdays while the schedule established. Turn to the 2nd step.
weekends and night shifts are for closing the day. (viii) If all the bees have been formed, then stop.
Therefore; scheduling these nurses as required is After producing the initial solution, the suitability value
achieved with the exact constraints of number four or of every bee is calculated and then the bees are
seven. Similar to this, situations at which the nurses arranged from small to large ones based on their
want (or don’t want) to work are suitably achieved with suitability values. The best bees are selected up to the
the fourth, seventh and ninth constraints. The types of number of the scout bees, and then the rest bees are
shifts applied for every shift are also shown on Table 3. erased. On Table 2, exact and soft constraints and their
To make sure that the assignments made are compatible respective punishment weights are shown.

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
128
 Bee Colony for Nurse Scheduling
Tablo 2. Constraints and penalty points.
Constrain
Weight Penalty function Violation measurement factor
No Hard Constrains

1 Every nurse must comply with the specified minimum working constraint 1000 exponentially = total work – minimum necessary working time

A person can work successively at maximum of 16 hours; no one should work

2 1000 exponentially = number of successive work - 2
continuously for 24 hours.
= 1; if no rest is made,
3 There must be at least one resting period of 48 successive hours per week. 750 linearly
0; otherwise
= 1; if no work was done on the required shift,
4 Nurses are obliged to work on the days and shifts they are assigned for. 1000 linearly
0; otherwise
= 1; if working was done for 16 hours on two consecutive days,
5 No working for 16 hours within two consecutive days. 500 linearly
0; otherwise
= 3 + (3 – number of successive rest); if rested less than 24 hours,
6 There must be a resting period of at least 24 hours after a 16-hour working shift. 150 exponentially
0; otherwise
= 1; if no resting was achieved in the required shift,
7 Efforts must be made to meet the nurses’ working leaves. 750 linearly
0; otherwise
One should work on one shift only within a day; working on two separate shifts is
8 500 linearly = 1; if rested for less than 16 hours, 0; otherwise
prohibited.
The tendency that nurses are sometimes, not in their working moods should be taken = 1; if no rest was made at required shift,
9 500 linearly
into account. 0; otherwise
= 1; if work was done beyond the specified shifts,
10 Works must be carried out at their specified shift types. 500 linearly
0; otherwise
Constrain
Soft Constrains
No
= 6 – number of successive rests; if rested less than 48 hours,
11 It will be better if a 48-hour resting period is offered after 16 working hours. 150 exponentially
0; otherwise

12 Fair distribution of overtimes among the workers is recommended. 100 exponentially = number of overtimes made

13 Equalization of workers’ night shift periods is needed. 40 exponentially = number of night shifts made

= 1; if working was also done on the two days of weekends,

14 Same nurse must never work two days on weekends. 50 exponentially
0; otherwise

15 If possible, nurses should work on successive days. 100 linearly = number of non-successive holidays

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
129
 K. Buyukozkan, A. Sarucan

Table 3. Department details and special cases.

Planning Number of Shift type

Department Nurse requests and preferences
period nurses
8*16, 1st nurse, the responsible nurse
Plastic 05 March - 01 16*8, 2nd nurse will be on duty on Saturdays of the 1st and 2nd week
7
Surgery April 2012 8*24,
16*24
1st nurse, the responsible nurse
Obstetrics 05 November - 4th nurse wants to work at night in the weekdays and at day times on weekends
8*16,
and 02 December 6 3rd nurse wants to work on Saturday in the 1st week, on Wednesday and Sunday in
16*8
Gynecology 2012 the 2nd week, on Friday in the 3rd week, on Tuesday and Sunday in the 4th week at
16*8 shifts.
1st nurse, the responsible nurse
2nd nurse will work on day shifts in the first three weeks, and at 16*8 shift on
01 October -28 8*16, Friday in the 4th week
Pediatrics 7 2nd nurse is off duty on Wednesday, Thursday and Friday of the 1st week.
October 2012 16*8
4th week, all workers will work 16 hours less than normal due to festival holiday
6th nurse is off duty until Saturday of the 4th week.
1st nurse, the responsible nurse
2nd, 3rd and 5th nurses shall work on day times during weekdays
8*16, 2nd nurse is on the 4th week leave.
03 September -
16*8, 6th nurse spends her annual leave throughout four-week period
Oncology 30 September 13
8*24,
2012 8th nurse is on her first two weeks of the annual leave
16*24
11th nurse is on annual leave from Monday of the second week up to Wednesday of
the 3rd week.
12th nurse is on the third week of her annual leave
1st nurse, the responsible nurse
2nd nurse wants to work twice at the 16*8 shift on weekdays
3rd nurse is on annual leave with the exception of Wednesday of the second week
7th nurse works 32 hours a week and can only work at the 8*16 shift.
8*16, 8th nurse wants to work on Monday and Tuesday in the third week
06 August - 02
Eye 16*8, 8th nurse is on her last week of her annual leave.
September 11
Treatment 8*24,
2012 9th nurse is on the third and fourth week of her annual leave.
16*24
10th nurse is the first and second week of her annual leave .
Because Monday and Tuesday of the 3rd week fall within festival period, all
workers shall work 16 hours less than normal.
Because Thursday of the 4th week is an official holiday, all workers shall work 8
hours less than normal times.

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
130
 Bee Colony for Nurse Scheduling

Table 4. Nurse demands for the departments.

First Week Second Week Third Weed Fourth Week

Department\Days
Mo Tu We Th Fr Sa Su Mo Tu We Th Fr Sa Su Mo Tu We Th Fr Sa Su Mo Tu We Th Fr Sa Su
3 3 3 3 3 1 1 3 3 3 3 3 1 1 3 3 3 3 3 1 1 3 3 3 3 3 1 1
Plastic Surgery 2 2 2 1 2 1 1 1 2 2 2 2 1 1 1 2 2 2 2 1 1 1 2 2 2 2 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
3 3 3 2 3 1 1 3 3 3 3 2 1 1 3 3 3 3 2 1 1 3 3 3 3 2 1 1
Obstetrics and
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
Gynecology
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
3 3 2 2 2 1 1 3 3 3 3 2 1 1 3 3 3 3 2 1 1 3 3 1 1 2 1 1
Pediatrics 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
5 5 4 4 4 1 1 4 4 5 5 4 1 1 4 5 4 4 4 1 2 4 4 5 4 4 2 3
Oncology 2 2 3 3 3 2 2 2 2 2 2 3 2 2 2 2 2 2 2 2 2 2 2 3 2 3 2 2
2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2 2
3 3 3 3 3 2 2 4 3 4 4 3 2 1 1 1 4 3 3 3 2 3 3 3 2 3 1 1
Eye Treatment 2 2 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 3 2 2 3 2 1 1 1 1 1 1
2 2 1 1 2 1 2 2 2 2 2 2 1 1 1 1 1 3 1 2 2 1 1 1 1 1 1 1

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
131
K. Buyukozkan, A. Sarucan

Table 5. Parameters used for the recommended ABC algorithm and their solution periods parameters.

Number Number of CPU Time (second)

Number of
of Scout Follower Iteration First
Initial bee Second Week Third Weed Fourth Week
bees Bees Week
Plastic Surgery 150 15 20 1200 23,68 25,59 25,63 26,24
Obstetrics and Gynecology 150 15 20 1200 24,98 25,44 25,75 25,55
Pediatrics 150 15 20 1200 25,02 26,86 26,73 26,59
Department

Eye Treatment 150 15 20 1200 26,8 28,98 28,72 27,72

Oncology 150 15 20 1200 28,2 30,22 30,68 31,1

In case of a situation that necessitates penalty, then by After determining bee followers, the bees are compared
taking the data regarding that situation from Table 2, with respective follower bees and the best bee is chosen
penalty value is calculated and added onto the bee’s as the pioneer bee. After specifying every pioneer bee
penalty. in this manner, iteration becomes accomplished. Then
Penalty points are calculated in the same way both at the loop starts again and the algorithm runs until the
the initial solution and during evaluation of the follower required number of iterations is reached. The stopping
bees. The penalties of the nurses throughout a weekly condition in this study has been defined as the number
schedule are summed up to obtain the bees’ or of iterations. In order to determine the correct number
followers’ penalties. of iterations, the program was run several times with
Generally, the schedule obtained from the initial different types of problems.
solution becomes in a worse condition. To improve The model formed has been designed basically for
such a schedule, neighbor finding is applied around the establishing weekly schedules. When the program is
pioneer bees in order to establish the followers. run once, the first week solution will have been
Neighbor finding is achieved with the following established. Beside this, the model forms a four week
algorithm steps; schedule by taking the previous weeks’ schedules into
(i) Turn up to the number of pioneer bees consideration. In this way, schedules for a four-week
(ii) Equalize all the follower bees to the pioneer bees period will have correctly been formed.
(iii) Reset the counter
(iv) Turn up to the number of follower bees 4.2. The software developed for the problem
(v) Select two nurses randomly To achieve solution to the nurse scheduling problem a
(vi) Make one increment to the counter user interactive program was prepared on Visual Studio
(vii) Randomly choose a day C # software27. A useful user guide interface that
(viii) Select a shift randomly enables the program user to enter necessary data was
(ix) Randomly select two nurses formed. While most studies in literature are only
(x) Change the selected shifts for the selected nurses, capable of offering solutions to certain specific shift
calculate the follower penalty models, the developed program in this study allows the
(xi) If the new status is better or the counter = 5, then user to select any required shift type from the user
save the follower and increase the number of follower interface. In addition, the program has been designed in
by one. Otherwise return to step 6. such a way that, through its interface, it allows entrance
(xii) If the number of followers is reached, increase of penalty values for details such as the number of
the number of pioneer by one and return to the 2nd step personnel, demand values, exact and flexible
(xiii) If the number of pioneer bees is reached, then constraints. With these features, the developed program
compare every pioneer bee with its follower bee; provides a good opportunity for making changes, easily,
designate a good follower bee as a pioneer bee. to the characteristic specifications of the problem. This

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
132
 Bee Colony for Nurse Scheduling

flexible feature of the program provides a room for observed that the available number of workers is quite
applications with different hospitals and a diverse range sufficient to cover all the demands without necessitating
of the departments at those hospitals. them to work on overtimes. When the values for the
At Karadeniz Technical University, Faculty of plastic surgery department in Table 6 are studied, it is
Medicine, where this study was conducted, the seen that the proposed algorithm sufficiently covers all
scheduling programs are carried out by responsible the exact constraints under the given operating
personnel of every department. This forms a structure conditions and that 22 flexible constraints are infringed.
that enables more accurate and quick decision on Corresponding to this, when the values for plastic
workers special demands like workers leaves, types of surgery given in Table 7 are investigated, it is found
works etc. With the help of its flexible feature, the that, in the existing application, the third exact
study enables separate formation of schedules for every constraint was infringed twice whereas the sixth exact
department and simplifies their control by the constraint was infringed once. Besides this, flexible
responsible personnel of these departments. constraints were found to be infringed eighty times.
This proves that the proposed ABC algorithm model
4.3. Program print outs and evaluation exhibits better performance than the conventional
The parameters and solution periods used for the ABC method provided that extreme conditions are not the
algorithm conducted for the hospital departments are case.
given in Table 5. In order to determine suitable Investigation on values given for the second department
parametric values, the program was run several times. in Table 3 reveals that this department possesses more
When the table (Table 5) is studied it is found that same constraining special cases than the first department
parameter values are sufficient for all the departments. does. The fact that there are only two working shifts in
This shows that the parameters used were chosen the department plus inclusion of special demands of the
correctly. When the processing time is investigated it is nurses in the schedule cause slight toughening of the
seen that the program used between 23 and 31 seconds department. Table 4 also indicates that the existing
to establish a weekly schedule. Such a short processing number of personnel in the second department is
time increases usability of the program in real life. capable of covering the nurses’ demands without
The difference between the longest and shortest having overtimes. As for the case of infringed
processing time for a weekly schedule is about 8 constraints listed in Table 6, it is seen that the proposed
seconds. This shows how insignificant the effects of algorithm has had only one exact constraint infringed
tough conditions or increasing the number of personnel while flexible constraints have been infringed twenty
may be to the program. The schedules obtained from four times. Table 7 shows the values for the second
the conducted studies and the existing schedules were department where by the existing method has
evaluated based on specified constraints. Table 6 shows undergone exact constraint infringement four times and
the constraints infringing numbers for every department flexible constraints sixty five times.
calculated by the developed ABC program. These As for the remaining 3rd, 4th and 5th departments, when
constraints’ infringing numbers are the total of values Table 3 is studied it is found that special cases are
for a four-week period. As for the schedules applied by progressively increasing and that scheduling procedures
the hospital management, the constraint infringing become more complex. Table 6, shows that even under
numbers are given in Table. these tough conditions the proposed algorithm still
When Table 3 is studied, it is seen that there is no exhibits better performance by maintaining all the exact
special constraining conditions, whatsoever, for the constraints for the 3rd and 5th departments but infringing
plastic surgery department. Higher number of shifts four exact constraints for the 4th department. On the
worked at, simplifies the assigning process. Within the contrary, when Table 7 is investigated, it is found that
planning period, none of the personnel took a leave. If in the applied schedule 3, 16 and 5 exact constraints
the demand values given in Table 4 are studied, it is were infringed in the 3rd, 4th and 5th departments
respectively. By comparing these two situations, it is

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
133
K. Buyukozkan, A. Sarucan

clearly seen that the proposed algorithm forms more infringed flexible constraints is 208; but the total
quality schedules even under difficult conditions than number of exact constraints infringed in the applied
the existing schedule. The result tables (Table 6 and schedules (Table 7) is 31 while that of the flexible
Table 7) show that the proposed algorithm is capable of constraints within that system is 457. These results
forming more improved schedules for all the five confirm that, under various operating conditions, the
departments as compared with the normally applied proposed algorithm runs effectively and establishes
system. As it is also seen from Table 6, the total of suitable nurse schedules in a quick manner.
infringed exact constraints is 5 and the total number of

Table 6. Number of infringed constraints in the schedules formed with the proposed ABC algorithm.

Hard Constraints (1, 2,…,10) Soft Constraints (11, 12,…,15)

No Department
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 Plastic Surgery 0 0 0 0 0 0 0 0 0 0 0 0 6 4 12
Obstetrics and
2 0 0 0 0 0 0 0 0 1 0 3 0 5 1 15
Gynecology
3 Pediatrics 0 0 0 0 0 0 0 0 0 0 0 2 4 4 16
4 Eye Treatment 0 0 0 0 0 0 1 1 0 2 15 3 9 6 23
5 Oncology 0 0 0 0 0 0 0 0 0 0 21 4 11 5 34

Table 7. Number of infringed constraints in the schedules prepared by the hospital management.

Hard Constraints (1, 2,…,10) Soft Constraints (11, 12,…,15)

No Department
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15
1 Plastic Surgery 0 0 2 0 0 1 0 0 0 0 39 0 6 4 31
Obstetrics and
2 0 0 1 0 0 0 0 0 3 0 33 0 8 3 21
Gynecology
3 Pediatrics 0 0 3 0 0 0 0 0 0 0 24 4 4 3 20
4 Eye Treatment 0 0 2 0 3 3 0 0 4 4 60 3 11 7 33
5 Oncology 0 0 1 0 2 2 0 0 0 0 83 4 15 6 35

results obtained from the algorithm were compared with

5. Results and Suggestions the existing applied method.
In this study, Artificial Bee Colony (ABC) algorithm The constraints and penalty points considered during
has been proposed for the Nurse Scheduling Problems scheduling, literature studies and face to face contacts
(NSP) which is termed as a NP-difficult problem in made with hospital personnel were specified based on
literature. In order for the proposed algorithm to work a the real situations. The obtained results indicated that
user interface model was developed on a PC by using a algorithm exhibits improved performance under all
flexible C# programming language. conditions by substantially covering both exact and
This is the first study that uses ABC algorithm to deal flexible constraints, forming more improved schedules
with the NSP. In order to evaluate the performance of than the existing method and that the new method is
the proposed ABC algorithm, real data obtained from faster too. Consequently; this study has shown that
five departments of the hospital belonging to Karadeniz ABC algorithm is a suitable method for NSP.
Technical University, Faculty of Medicine were used. Future studies may include parameter analysis and
Shift schedules were formed by using the data and the hybrid methodologies to speed up the model even
further and enhancing the solution quality. Moreover;
the same data may be used and the proposed algorithm

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
134
 Bee Colony for Nurse Scheduling

may be compared and contrasted with other intuitive International Transactions in Operational Research, 10
methods like Genetic Algorithm, Variable Neighbor (2003) 245–265.
14. M. V. Chiaramonte and L. M. Chiaramonte, An agent-
Search, and Tabu Search
based nurse rostering system under minimal staffing
conditions, International Journal Production Economics,
6. References 114(2) (2008) 697–713.
1. C. Valouxis and E. Housos, Hybrid optimization 15. B. Maenhout and M. Vanhoucke, An evolutionary
techniques for the workshift and rest assignment of approach for the nurse rerostering problem, Computers
nursing personnel, Artificial Intelligence in Medicine, and Operations Research, 38(10) (2011) 1400-1411.
20(2) (2000) 155–175. 16. M. Cheng, H. I. Ozaku, N. Kuwahara, K. Kogure and J.
2. E. Burke, P. Cowling, P. D. Causmaecker and G. V. Ota, Simulated annealing algorithm for daily nursing
Berghe, A memetic approach to the nurse rostering care scheduling problem, Proc. 3rd Annual IEEE
problem, Applied Intelligence, 15(3) (2001) 199–214. Conference on Automation Science and Engineering,
3. E.K. Burke, T. Curtois, G. Post, R. Qu and B. Veltman, (Scottsdale, AZ, USA, 2007), pp. 507–512.
A hybrid heuristic ordering and variable neighbourhood 17. D. Parr and J. M. Thompson, Solving the multi-objective
search for the nurse rostering problem, European J. of nurse scheduling problem with a weighted cost function,
Operational Research, 188(2) (2008) 330–341. Annals of Operations Research, 155(1) (2007) 279–288.
4. J. Li, E. K. Burke, T. Curtois, S. Petrovic and R. Qu, The 18. B. Maenhout and M. Vanhoucke, An electromagnetic
falling tide algorithm: a new multi objective approach for meta-heuristic for the nurse scheduling problem, Journal
complex workforce scheduling, Omega-Int. J. of Man. of Heuristics, 13(4) (2007) 359–385.
Science, 40(3) (2012) 283–293. 19. P. Brucker, E. K. Burke, T. Curtois, R. Qu and G. V.
5. U. Aickelin and K. A. Dowsland, An indirect genetic Berghe, A shift sequence based approach for nurse
algorithm for a nurse-scheduling problem, Computers scheduling and a new Benchmark dataset, Journal of
and Operations Research, 31(5) (2004) 761–778. Heuristics, 16(4) (2010) 559–573.
6. F. Bellanti, G. Carello, F. Della Croce and R. Tadei, A 20. B. Maenhout and M. Vanhoucke, Comparison and
greedy-based neighborhood search approach to a nurse hybridization of crossover operators for the nurse
rostering problem, European Journal of Operational scheduling problem, Annals of Operations Research,
Research, 153 (2004) 28–40. 159(1) (2008) 333–353.
7. E. K. Burke, P. Causmaecker, S. Petrovic and G. V. 21. C. Tsai and S. H. A. Li, A two-stage modeling with
Berghe, Metaheuristics for handling time interval genetic algorithms for the nurse scheduling problem,
coverage constraints in nurse scheduling, Applied Expert Systems with Applications, 36 (2009) 9506–9512.
Artificial Intelligence, 20(9) (2006) 743–766. 22. A. Oughalime, W. R. İsmail and L. C. Yeun, A tabu
8. W. J. Gutjahra and M. S. Raunerb, An ACO algorithm search approach to the nurse scheduling problem,
for a dynamic regional nurse-scheduling problem in International Symposium on Information Technology, 1
Austria, Computers and Operations Research, 34 (2007) (2008) pp. 1–7.
642–666. 23. B. Cheang, H. Li, A. Lim and B. Rodrigues, Nurse
9. G. Beddoe, S. Petrovic and J. Li, A hybrid metaheuristic rostering problems-a bibliographic survey, European
case-based reasoning system for nurse rostering, Journal Journal of Operational Research, 151 (2003) 447–460,
of Scheduling, 12(2) (2009) 99–119. 24. P. Causmaecker and G. V. Berghe, A categorisation of
10. H. Çivril, Genetic Algorithms for Nurse Scheduling nurse rostering problems, J. Sched, 14(1) (2010) 3 – 16.
Problems, M.Sc. Thesis, (Süleyman Demirel University 25. A. Baykasoğlu, L. Özbakır and P. Tapkan, Swarm
Graduate School of Applied and Natural Science, Isparta, intelligence: focus on ant and particle swarm
Turkey, 2009). (Language: Turkish). optimization, Edited by Felix T. S. Chan and Manoj
11. B. Bilgin, P. Causmaecker, B. Rossie and G. V. Berghe, Kumar Tiwari, I-Tech Education and Publishing, 1th
Local search neighbourhoods for dealing with a novel edn. (Vienna, Austria, 2007), pp. 114–144.
nurse rostering model, Annals of Operations Research, 26. T. D. Seeley, S. Camazine and J. Sneyd, Collective
194(1) (2012) 33–57. decision-making in honey bees: how colonies choose
12. E. K. Burke, J. Li and R. Qu, A hybrid model of integer among nectar sources, Behav Ecol Sociobiol, 28 (1991)
programming and variable neighbourhood search for 277–290.
highly-constrained nurse rostering problems, European 27. K. Buyukozkan, A Bee Colony Algorithm For Nurse
J. of Operational Research, 203(2) (2010) 484–493. Scheduling Problem And An Application In A Health
13. T. M. Dias, D. F. Ferber, C. C. Souza and A. V. Moura, Service System, M.Sc. Thesis, (Selçuk University,
Constructing nurse schedules at large hospitals, Institute of Natural Sciences, Konya, Turkey, 2012).
(Language: Turkish).

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
135
K. Buyukozkan, A. Sarucan

28. S. Nakrani and C. Tovey, On honey bees and dynamic

server allocation in internet hosting centers, International
Society for Adaptive Behavior, 12(3–4) (2004) 223–240.
29. X. Yang, Engineering optimizations via nature-inspired
virtual bee algorithms, IWINAC, (Berlin Heidelberg,
2005), pp. 317–323.
30. S. Dehuri, S. Cho and A. K. Jagadev, A multi-agent
approach for multiple campaigns assignment problem,
International Conference on Information Technology,
(2008) pp. 24–29.
31. P. Tapkan, L. Özbakır and A. Baykasoğlu, Bees
algorithm and generalized assignment problem:
comparison of different neighborhood structures, J.
Industrial Engineering, 21(2) (2008) 2–13. (in Turkish).
32. C. Lara, J. J. Flores and F. Calderon, Solving a school
timetabling problem using a bee algorithm, MICAI,
(Berlin Heidelberg, 2008), pp. 664–674.
33. T. Davidovic, M. Selmic and D. Teodorovic, Scheduling
independent tasks: bee colony optimization approach,
17th Mediterranean Conference on Control &
Automation, (Makedonia Palace, Thessaloniki, Greece,
2009), pp. 1020–1025.
34. L. Özbakır, A. Baykasoğlu and P. Tapkan, Bees
algorithm for generalized assignment problem, Applied
Mathematics and Computation, 215(11) (2010) 3782 –
3795.
35. M. H. Kashani, M. Jamei, M. Akbari and R. M. Tayebi,
Utilizing bee colony to solve task scheduling problem in
distributed systems, Third International Conference on
Computational Intelligence, Communication Systems and
Network, (2011) pp. 298–303.
36. A. Kaur and S. Goyal, A Survey on the applications of
bee colony optimization techniques, Int. J. Computer
Science Eng., 3(8) (2011) 3037.
37. D. Teodorovic, T. Davidovic and M. Selmic, Bee colony
optimization: the applications survey, Transactions on
Computational Logic, (2011) 1–20.
38. B. Suri and Snehlata, Review of artificial bee colony
algorithm to software testing, IJRRCS, 2(3) (2011) 706–
711.

Co-published by Atlantis Press and Taylor & Francis

Copyright: the authors
136