Solving University/Polytechnics Exam Timetable Problem

using Particle Swarm Optimization

-"
" -
&
#*&'(
)!
)"%)&
 !(&!
#
%(
$!( #

!#
)!(#
'"! )!(#
!)#

)!

--


)!
)"%)& &'&#
'+#
!,'  ',#



!"
-"" )# !)", '- %')",

ABSTRACT insufficient of facilities like classes, rooms, seminar hall and

The timetabling problem at universities is an NP-hard problem lecturers. Particle swarm optimization (PSO) [1, 2, 3] is an
under multiple constraints and limited resources. Thus a technique alternative population-based evolutionary computation technique
that can handle constraints is needed to optimize the problem. In developed by Dr. Eberhart and Dr. Kennedy in 1995, inspired by
this paper we proposed an Exam Timetable generator focusing on social behavior of bird flocking or fish schooling. PSO shares
Particle Swarm Optimization (PSO) for finding optimal solutions many similarities with evolutionary computation techniques such
to the problem at Universiti Kuala Lumpur, MIIT faculty and as Genetic Algorithms (GA) [6, 7, 8]. Unlike GA, PSO has no
local institution. PSO is a promising scheme for solving NP - hard evolution operators such as crossover and mutation. Therefore,
problems due to its fast convergence and fewer parameter settings. PSO attempts to simulate social behavior, which differs from the
There are two objectives in this. First provide a detailed natural selection schemes of genetic algorithms. In PSO, the
introduction to the topic of timetabling, Particle Swarm population consists of potential solutions, called particles, similar
Optimization their method and their variations. The second to birds in a flock. PSO is the particles. As the particle moves it
objective is to apply them to the problem of Exam Timetabling. will updates its velocity and position based on its own best
The proposed algorithm will be tested using the timetabling data experience and that of the entire population. The updating policy
from UniKL MIIT City Campus Kuala Lumpur as well as will cause the particle swarm to move toward a region with a
Politeknik Sultan Salahudin Abdul Aziz Shah, Shah Alam. higher object value.

Categories and Subject Descriptors

G.4 [Mathematics of Computing]: MATHEMATICAL 2. PROBLEM DESCRIPTION
SOFTWARE: Algorithm design and analysis, Certification and In this specific timetable scheduling problem, the events are
testing, H.2.8 [Information Systems]: Database Applications: exams, and the times are the separate time slots. There are three
Data mining days for the allocation of the exams. Each day contains two time
slots, one is in the morning and another one is in the afternoon.
General Terms Table 1 illustrates the exam timetable framework used for this
Algorithms, Management, Measurement, Documentation, paper, where 1, 2, 3, ···, 22 are time slots. A student may take an
Performance, Design, Experimentation. average of 6 exams because they may take different number of
courses. Classes population at UniKL may varies from 10 to 90
students while classes in Politeknik varies from 40 to 50 students
Keywords
University Timetabling, Particle Swarm Optimization. Timeslots requirement for examination can be calculated from the
number of subject with exam offered multiply with the number of
1. INTRODUCTION courses and semester.
Classes in Universiti Kuala Lumpur (UniKL) MIIT and
Polytechnics Institution are getting bigger in size. As student For example:
intake and courses offered are increasing each year, a better
management in scheduling should be considered. An exam A Diploma course with five (5) semester levels and six (6)
timetabling application which is more flexible in generating the subjects offered per semester level would require 30 timeslots for
schedule is critical when there are many constraints including examination. For three (3) diploma course that would increase to
90.

© 2016 Association for Computing Machinery. ACM acknowledges that A Degree course with seven (7) semester levels and six (6)
this contribution was authored or co-authored by an employee, subjects offered per semester level would require 42 timeslots for
contractor or affiliate of a national government. As such, the examination. For three (3) degree courses that would increase to
Government retains a nonexclusive, royalty-free right to publish or 126.
reproduce this article, or to allow others to do so, for Government
purposes only.
Thus total timeslots required for 3 Diploma and 3 Degree courses
IMCOM '16, January 04-06, 2016, Danang, Viet Nam
© 2016 ACM. ISBN 978-1-4503-4142-4/16/01…$15.00
is at maximum 216 timeslot.
DOI: http://dx.doi.org/10.1145/2857546.2857593
 best (Pbest) and own information. Each particle will also
With two (2) timeslots per day and 10 rooms for examination, exchange its information to other particles and remember its
there will be a total of 20 examinations per day. For 216 global best (Gbest). Each particle has fitness value. Fitness value
examination, the number of day requirement would be 216/20 ~ of each particle’s current position is compared with Pbest and
11 days or 22 timeslots. (Monday thru Saturday and Monday thru Gbest. If it is better than Pbest and Gbest then update the Pbest
Friday) and Gbest. Equations (1,2) as shown below are applied to update
Table 1 Examination Timeslots the position and velocity of each particle. Figure 1 show the flow
Day\Time 9.00-12.00 AM 2.00-5.00 PM chart of standard PSO.[4]

Day1 1 2 Vid=Vid + c1 * r1 * (Pid - Xid) + c2 * r2 *(Pgd - Xid) (1)

Day2 3 4
Xid=Xid+Vid (2)
:
:
In the equation
Day11 21 22
Vid is the velocity component of ith particle in the dth
dimension
A matrix of timeslots versus rooms is set as show in Table 2. For
every timeslots, populate each room with an examination paper.
Xid is the position component of ith particle in the dth
We need to consider for constraint that occur. The hard constraint
dimension
cannot be violated. No student should sit for more than one exam
at the same timeslot. If hard constraints not violated, exam
Pid is the position component of the Pbest of ith
schedule is ready and can be used, however it can be further
particle in dth dimension
optimized for effectiveness of the schedule.
Pgd is the position component of the Gbest in dth
Table 2 Examination Allocation dimension
Slot\Rooms Rm1 Rm2 .. .. Rm10
c1 and c2 are learning factors
Slot 1 Ex1 Ex5
Slot 2 r1 and r2 are random numbers between [0,1]
: Ex6 Ex2
: Ex3
Slot 22 Ex4
3. PROBLEM FORMULATION
3.1 Resources and Initial State Values
2.1 Constraints In this specific timetable scheduling problem, it is assumed that,
The most important constraint of the exam timetable is that the the resources are defined as below:
same student cannot take two different exams at the same time. In
other words, the exams cannot clash for any student. This is the T a finite set of timeslots
only hard constraint. R a finite set of resources/rooms
However, we also consider the several soft constrains: X a finite set of exam
• Very strongly prefer no two exams per day for any student. S a finite set of students registered for exam
• Strongly prefer student to have a least an empty slot in D a finite set of day
between exam. Z a finite set of exam paper offered
C a finite set of constraints
2.2 Penalty for Fitness Calculation
To determine the best placement of examination while following
the constraint requirement, penalty will be calculated which will We assumed initial values of resources as follows:
result in reduced fitness for a particular population.
In order calculate penalty, if examination for a particular student T approximately 220 slots
fall consecutively which didn’t meet the soft constraint (2 timeslot/day * 11 days * 10 rooms)
requirement, the particle will be penalize. For best result, R 10 rooms
examination has to be schedule with a minimum of one (1) empty S 50 student/course * 6 course/sem * 6
timeslot in between exam. This is called the proximity cost [5]. subject/student = 1800
2.3 Particle Swarm Optimization (PSO) Z 216 exams-paper (6 courses * 6
Eberhart and Kennedy developed Particle swarm optimization subjects/course * 6 semester)
algorithm in 1995. It is a technique based on particles, each D 11 days
particle has velocity in search space in order to find feasible The problem is to assign timeslots and rooms for the exam as to
solution and adjust its position according its own previous satisfy the constraints as far as possible. The most important
experience and neighbors’ experiences. The initial position and constraint of the exam timetabling is that the same student cannot
velocity of each particle is anomaly determined. When each take two different exams at the same time. In other words, the
particle to move a new position then it will remember its personal exams cannot clash for any student. This is the only hard
constraint. However there are also several soft constraints: i) Very 4. Completing the PSO procedure will follow with another
strongly prefer no two exams per day for any student. ii) Strongly fitness evaluation session.
prefer student to have a least an empty slot in between exam.
5. The PSO procedure will then continue for a number of
generations.

The final result will be obtain from the population with the best
fitness values.

Figure 1 Standard PSO flowchart

3.2 Propose PSO Procedure

The following steps are the proposed procedure for the
examination timetabling generator.
1. Initial population of examination is created by assigning
examination to timeslots. All examination will be
allocated onto the Timeslot vs Rooms matrix as shown
in Table 2.
2. The fitness of this population will be calculated using
formula to calculate the proximity between
examinations for student. A proximity cost x is given
whenever a student has to sit for two examinations
scheduled s periods apart: these weights are x1 = 8, x2 =
4, x3 = 2 and x4 = 1. The higher the cost the lower the
fitness. Figure 2 Examination Timetable Flow with PSO

3. The population will go through the standard PSO

routine as shown in Figure 1.
3.3 Simulation and Experiment
The above method and algorithm is proposed to be used to solve [3] M. Gen and R. Cheng, Genetic Algorithm and Engineering
for examination timetabling problem in order to further optimize Design, John Wiley and Sons, New York, 1997.Tavel, P.
the current procedure at UniKL MIIT as well as at the Politeknik 2007. Modeling and Simulation Design. AK Peters Ltd.,
Sultan Salahudin Abdul Aziz, Shah Alam. Natick, MA.
[4] Shu-Chuan Chu, Yi-Tin Chen; Jiun-Huei Ho, Timetable
4. ACKNOWLEDGMENTS Scheduling Using Particle Swarm Optimization. In
We would like to thanks UniKL MIIT for all the opportunities Proceedings of the First International Conference on
given. Innovative Computing, Information and Control (ICICIC'06)
[5] Graham Kendall and Naimah Mohd Hussin, Tabu Seacrch
5. REFERENCES Hyper-heuristic Appraoch to the Examination Timetabling
[1] D. E. Goldberg, Genetic Algorithm in Search, Optimization Problem at University technology MARA, In the Proceeding
and Machine Learning, Addison-Wesley Publishing of The International Series of Conferences on the Practice
Company, 1989. and Theory of Automated Timetabling (PATAT2004)
[2] L. Davis, Handbook of Genetic Algorithms, Van Nostrand
Reinhold, New York, 1991.