Applied Mathematics and Computation 179 (2006) 702–706

www.elsevier.com/locate/amc

Improving paper spread in examination timetables

using integer programming
S.A. MirHassani
Department of Mathematics, Shahrood University of Technology Shahrood, P.O. Box 36155-316, Iran

Abstract

One of the desirable attributes of real-life examination timetabling solutions is the maximization of paper spread, which
is a measure of the amount of study time that each student has between examinations. In this study, we face with a pre-
defined examination schedule that must be modified in order to maximize paper spread in course of examination. We will
show how the integer programming can be employed to achieve this aim. The model presents constraints for the quality of
feasible examination timetables and all the requirements found in most academic institutions. The approach is tested on
real-world exam timetabling problems. The computational experiments and results will be reported.
 2005 Published by Elsevier Inc.

Keywords: Examination timetabling; Assignment; Integer programming

1. Introduction

Timetabling problems consist of the feasible assignment of time-slots to a set of events, subject to a set of
constraints in such a way that a set of desirable objectives will be satisfied as much as possible. Timetabling is a
difficult (NP-complete) problem [1] in the general case, except when the constraint matrix is, for example,
totally unimodular (rarely in real life) and belongs to a general class of problems known as scheduling.
Due to the variety of constraints typical in different timetabling environments, it has been difficult to develop
a generic solution to the problem of timetabling. There exist a growing number of solutions to the different
types of timetabling problems with different types of constraints. Since there is no common standard on spec-
ifying a timetabling problem instance and its solution proposed by a researcher, most of the results cannot be
compared and benchmarking thus becomes almost impossible. Consequently, examination timetabling ren-
ders a difficult combinatorial optimization problem and too complex to be resolved via manual means. Appro-
priate algorithms are required to provide adequate examination timetables for universities.
The difficulty of developing appropriate examination timetables for educational institutions is ever increas-
ing. Institutions are enrolling more students into a wider variety of courses including an increasing number of
combined degree courses. So the difficulty of developing appropriate examination timetables for educational

E-mail address: [email protected]

0096-3003/$ - see front matter  2005 Published by Elsevier Inc.

doi:10.1016/j.amc.2005.11.125
 S.A. MirHassani / Applied Mathematics and Computation 179 (2006) 702–706 703

institutions is demanding. Timetabling problems have been attracting the attention of the scientific research
community across Artificial Intelligence and Operational Research for more than 40 years [2,3]. Schaerf [4]
studied problems and classified them as high school timetabling, university course timetabling and examina-
tion timetabling problems (ETP). Merlot et al. [5] provides a hybrid algorithm for the examination timetabling
that consists of three phases including constrain programming. In the early days of timetabling research, graph
heuristics [6] were widely studied. They represent simple techniques but tend to be impractical for complex
problems (at least when implemented on their own). Integer linear programming is an exact solution method
which tends to be computationally very expensive specifically in case of solving large timetabling problems.
Papoutsis and et al. [9] provide a column generation approach for the solution of timetabling problems. White
et al. [10] used Tabu search to create examination timetable.
A general timetabling problem includes the assignment of a set of events (exams, courses, sports matches,
meetings, etc.) into a limited number of timeslots (time periods), while satisfying a set of constraints. These
constraints are usually grouped into two types, which are described below

• Hard constraints which cannot be violated under any circumstances. For example, a person cannot be
assigned to two different events at the same time. Solutions which do not violate any of the hard constraints
are called feasible solutions.
• Soft constraints are desirable but not essential. In most real world situations no solution can be found
which satisfies all of the stipulated soft constraints.

A typical real-life examination-timetabling problem involves both critical (hard) and non-critical (soft) con-
straints. The satisfaction of all critical constraints determines if a given schedule is feasible. The quality of a
feasible timetable, however, is determined by some weighted sum of the satisfaction of non-critical constraints.
One such non-critical constraints that exists in real-life examination timetables is paper spread or exam prox-
imity, which is a measure of the amount of study time that each student has between examinations. In the
exam proximity problem, a timetable has to offer a student’s maximum free time between exams while satis-
fying the clashing constraint (exam conflicts) with no regard as to the sitting capacity.
The structure of the paper is simple. The first section briefly discusses the problem background and its char-
acteristics in general. The second section presents the description of an examination timetabling problem. The
third section presents the mathematical programming model. The paper ends with experimental results and a
brief conclusion.

2. Problem description

The third class examination timetabling, which is our concern has the following characteristics:

• There is only one exam for each subject.

• The conflicting condition is generally strict. For example, a student can skip a lecture due to an overlapping
in the program, but the student cannot skip an exam.
• There are different types of constraints, for example, at most one exam per day for each student, and not
too many consecutive exams for each student.
• The number of periods may vary, in contrast to course timetabling where it is fixed.
• There can be more than one exam per room.

There are a number of Courses and one Exam for each Course. There are some groups of Exams, such that
each Exam group consists of students taking a number of Exams.
In our case, there is a predefined examination timetable which is based on current course’s schedule. Orig-
inally, the timetable is feasible for all available courses according to their setting in the semester timetable
which all the students have to observe as mach as possible. Since some students will be graduated at the
end of semester, they have to enroll for all the remaining courses which they need for graduation. Under these
circumstances, the available examination timetable will be infeasible and would need to be modified as much
as possible in order to overcome such difficulties and hence to resolve as many of exam’s overlapping as pos-
704 S.A. MirHassani / Applied Mathematics and Computation 179 (2006) 702–706

sible. So a model is designed to fix a currently available timetable which is originally produced and provide
adequate examination timetables while spreading the exams for the students as much as possible to offer stu-
dents a better study time between examinations.

3. The model

In this work, we mainly try to formulate all the examination timetables’ requirements. To support maxi-
mum exam spreading, we consider three types of exam clashing that may happen for a student

(I) Two exams in the same time-slot.

(II) Two exams on the same day.
(III) Two exams on two consecutive days.

Students usually prefer conflicts of the type III to II and Type II to I exam clashing. Type I is a real type
conflict and therefore a hard constraint while the others can be considered as soft constraints. The objective
function is placed in a way as to minimize the number of different conflicts of type I, II, III.
Thus, we have the following model characteristics:

Sets
• A set S = {1, . . . , s} of students with different courses and s 2 S.
• A set C = {1, . . . , c} of courses and c 2 C.
• A set D = {1, . . . , d} of days and d 2 D.
• A set T = {1, . . . , t} of time-slot or session per day and t 2 T.
• Qs number of students with the same set of courses.

Data
• GSc The group of students that enrolled for course c.
• GCs The group of courses in which the students enrolled.

Variables
• Xcdt = 1 if the exam of course c, set on day d, session t, otherwise 0 (Binary variables).
• Ysdt = 1 if the student s, has an exam on day d, session t, otherwise 0 (Binary variables).
• Usdt = the number of exceeded courses exam that student s, has on day d time t. For example if
Psdt = 2 then this student must exam three course at the same time.
• Vsd = the number of exceeded course exams that student s, has on day d.
• Wsd = the number of exceeded course exams that student s, has on day d and the day after.

Objective
P function components
• P Qs U sdt The number of students with conflict type I.
• P Qs V sd The number of students with conflict type II.
• Qs W sd The number of students with conflict type III.

Model
X X X
Min Z ¼ K U sdt þ L Qs V sd þ M Qs W sd .
s;d;t s;d s;d

Through this objective function we panelize each conflict of type I, II, and III by fixed values K, L, and M,
respectively. In order to force priorities we set M  L  K.
Constraints
X
X cdt ¼ 1 8c. ð1Þ
dt
 S.A. MirHassani / Applied Mathematics and Computation 179 (2006) 702–706 705

Only one time-slots for course c exam

X cdt 6 Y sdt 8c; d; t 8s 2 GSc . ð2Þ

These constraints force the following condition that defines the relationship between exams and students. If
Xcdt = 1 then "s 2 GSc,Ysdt = 1 Otherwise Ysdt = 0
X
X cdt 6 U sdt þ 1 8s; d; t. ð3Þ
c2GCs

These constraints measure the number of conflicts of type I

X
Y sdt 6 V sd þ 1 8s; d. ð4Þ
t

These constraints measure the number of conflicts of type II

X X
Y sdt þ Y s;dþ1;t 6 W sd þ 1 8s; d. ð5Þ
t t

These constraints measure the number of conflicts of type III

U sdt 6 1 8s; d; t. ð6Þ

Since practically the number of conflicts of type I never go beyond one so we would consider the above upper
bound for this set of variables. Also all variables are supposed to be nonnegative.

4. Experimental results

We implemented the above exam timetabling program with Mathematical Programming Language (MPL)
software [8] experimented on a Pentium III 700 MHz PC. Typical run time on the real data set from Shahrood
University of Technology (about 324 exams per semester in 30 sessions, based on 4000 enrollments) equates to
less than 2 min for the constraint programming problem. The solution time was very sensitive to three param-
eters K, L, and M. We tried it with different values and in some situations it took too much time. Finally, we
found a set of values for K, L, and M that in a reasonable solution time lead to the zero value for objective
functions. This means that a conflict-free examination timetable that provides better study times for the stu-
dents has been found.

5. Conclusion

As it is well recognized, people in general are not interested in solving their optimization problems to opti-
mality or even close to optimality. They are more often interested in ‘‘good enough—soon enough—cheap
enough solutions to their problems [7]. We also think that good choices for the specific parts of each problem
are fundamental for the success of any search algorithm. Therefore, we have successfully used constraint pro-
gramming for the problem of modifying exam timetables with real data sets. In this work, we started with a
predefined exam schedule and tried to make it feasible enough for all the students. Less solution time would be
available, if we start with better schedules and balance between the solution quality and search time. In spite of
the shortcomings of the comparisons, the method still proves as a promising algorithm, among the best cur-
rently used ones for examination timetabling.

Acknowledgements

I would like to thank my students Mr. Erfani and Mr. Mokhtar for their assistance in solving this problem
who helped me in many ways.
706 S.A. MirHassani / Applied Mathematics and Computation 179 (2006) 702–706

References

[1] S. Even, A. Itai, A. Shamir, On the complexity of timetable and multicommodity flow problems, SIAM J. Comput. 5 (4) (1976) 691–
703.
[2] E. Burke, K. Jackson, J. Kingston, R. Weare, Automated timetabling: the state of the art, Comput. J. 40 (9) (1997) 565–571.
[3] E. Burke, P. Causmaecker (Eds.), Selected Papers from the Fourth International Conference on the Practice and Theory of
Automated Timetabling, Springer-Verlag, LNCS 2740, 2003.
[4] A. Schaerf, A survey of automated timetabling, Technical Report, CWI-Amesterdam, 1995.
[5] L.T.G. Merlot, N. Boland, B.D. Hughes, P.J. Stuckey, A Hybrid Algorithm for the Examination Timetabling Problem, in: E. Burke,
P. De Causmaecker (Eds.), Practice and Theory of Automated Timetabling, Fourth International Conference, LNCS 2740, Springer-
Verlag, Gent, Belgium, 2003, pp. 207–231.
[6] D. Brelaz, New methods to color the vertices of a graph, Commun. ACM 22 (4) (1979) 251–256.
[7] E. Burke, E. Hart, G. Kendall, J. Newall, P. Ross, S. Schulenburg, Hyper-heuristics: an emerging direction in modern search
technology, in: F. Glover, G. Kochenberger (Eds.), Handbook of Meta-Heuristics, Kluwer, 2003, pp. 457–474.
[8] MPL modeling software system, Maximal software Inc., 2000.
[9] K. Papoutsis, C. Valouxis, E. Housos, A column generation approach for the timetabling problem of Greek high schools, J. Oper.
Res. Soc. 54 (2003) 230–238.
[10] G.M. White, B.S. Xie, S. Zonjic, Using Tabu search with longer-term memory and relaxation to create examination timetables, Eur. J.
Oper. Res. 153 (1) (2004) 80–91.