View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by Scientific Repository

Genetic Algorithm for Scheduling Courses

Gregorius Satia Budhi, Kartika Gunadi, Denny Alexander Wibowo

Petra Christian University, Informatics Department

Siwalankerto 121-131, Surabaya, East Java, Indonesia
e-mail: [email protected], [email protected]

Abstract. In the university, college students must be register for their classes,
and there still many college student that was confused on how to make a good
classes schedule for themselves. Mainly because of many variables and consid-
erations to be made, for examples, they have to consider how hard the classes
they are going to take, and also, they still have to consider their exam schedules
and also their availability time as well. Genetic Algorithm is one of many meth-
ods that can be used to create a schedule. This method determines the best sched-
ule using fitness cost calculation which can compare the quality of one schedule
against the other. Then, using crossover, mutation, and elitism selections, we can
determine better schedules. Based on the result of the survey held before, 70% of
the respondents gave point 4 and 30% of the respondents gave point 5 out of 5
for the quality of the schedule made using this applications.

Keywords: Artificial Intelligence, Genetic Algorithm, Courses Schedule

1 Introduction

Schedule is very important to manage an activity, especially when the activity was car-
ried out in a large organization and held for long-term time or routine.
Within the scope of the university, the course schedule also is critical There are still
students who are confused when creating their own class schedules, because a lot of
consideration, for example: what subjects can be taken without clashing with another
schedule, a student who wants to the day - certain the day No lectures, due to personal
reasons or other student affairs, student who wants to take courses that would like to
correspond with their fields of specialization, and so forth. These things make them
have trouble making a personal course schedule. Through these problems, we conduct
research on a system that can help perform automatic scheduling lectures for students
by the day and the course desired by the student. The method used for the manufacture
of automatic schedule is Genetic Algorithm (GA). The reason is because GA is a
method that can find a solution that is both complex problems.

adfa, p. 1, 2011.
© Springer-Verlag Berlin Heidelberg 2011
2 Genetic Algorithm

Genetic Algorithm is a method to search an optimal solution for a problem. The method
will find a good solution by crossovers a possible solution with another solution to
create new solutions. After that method will mutate the new solutions so that they are
have parts of solution from the parents but not really same with the parent. The process
begin with the creation of random population of valid solutions / chromosomes, then
GA will count the fitness costs of each chromosomes in the population. After that two
chromosomes will be chose to crossover and produce offspring. Then the offspring will
be mutated. This process will be repeated till the stop condition was reached [1].

Given a clearly defined problem to be solved and a binary string representation for
candidate solutions, a basic GA can be represented as following major steps [2]:
1. Represent the problem variable domain as a chromosome of a fixed length, choose
the size of a chromosome population N, the crossover probability pc and the mutation
probability pm.
2. Step 2: Define a fitness function to measure the performance, or fitness, of an indi-
vidual chromosome in the problem domain. The fitness function establishes the basis
for selecting chromosomes that will be mated during reproduction.
3. Randomly generate an initial population of chromosomes of size N:

x1; x2; ...; xN (1)

4. Calculate the fitness of each individual chromosome:

f(x1) ; f(x2); ...; f(xN) (2)

5. Select a pair of chromosomes for mating from the current population. Parent chro-
mosomes are selected with a probability related to their fitness. Highly fit chromo-
somes have a higher probability of being selected for mating than less fit chromo-
somes.
6. Create a pair of offspring chromosomes by applying the genetic operators – crosso-
ver and mutation.
7. Place the created offspring chromosomes in the new population.
8. Repeat Step 5 until the size of the new chromosome population becomes equal to
the size of the initial population, N.
9. Replace the initial (parent) chromosome population with the new (offspring) popu-
lation.
10. Go to Step 4, and repeat the process until the termination criterion is satisfied.

2.1 Selection Criteria

The commonly used chromosome selection technique is the roulette wheel selection [3,
4]. It is like spinning a roulette wheel where each chromosome has a segment on the
wheel proportional to its fitness. The roulette wheel is spun, and when the arrow comes
to rest on one of the segments, the corresponding chromosome is selected. Each chro-
mosome is given a slice of a circular roulette wheel. The most fits chromosomes occupy
the largest areas, whereas the least fits have much smaller segments in the roulette
wheel. The area of the slice within the wheel is equal to the chromosome fitness ratio.
To select a chromosome for mating, a random number is generated in the interval (0,
100), and the chromosome whose segment spans the random number is selected [2].
For the example you could see in Table 1 and Figure 1.

Table 1. Example of Chromosomes and the fitness (Source: Negnevitsky, 2005).

Chromosome label Chromosome fitness Fitness ratio, %

X1 36 16.5
X2 44 20.2
X3 14 6.4
X4 14 6.4
X5 56 25.7
X6 54 24.8

Fig. 1. Example of the slices in roulette wheel selection (Source: Negnevitsky, 2005)

2.2 Reproduction
There are several methods used in the GA Reproduction one of them is crossover.
Crossover is the crosses of at least 2 solutions, where through parts crossed, will gen-
erate new solutions. Through this crosses GA will add solutions and variations that are
useful later in the scoring stage when searching for a better solution [5]. The example
of the crossover could be seen in Figure 2.
 Fig. 2. Crossover in binary chromosome (Source: Haupt, R.L., Haupt, S.E., 2004)

Besides the crossover, there is another way to complete reproduction, namely mutation.
Mutation is a method that converts a portion of the solution. Although altered a little
bit, it can create different variations resulting in different solutions and even new solu-
tion. One method used in this mutation is shuffle. [5]. Mutation role is to provide a
guarantee that the search algorithm is not trapped on a local optimum [6]. The example
of the mutation could be seen in Figure 3.

Fig. 3. Mutation in a binary chromosome (Source: Haupt, R.L., Haupt, S.E., 2004)

3 Courses Enrolling on Study Plan Registration time

Students will enroll subjects which wants to be taken at the beginning of the semester,
during the study plan registration (SPR/PRS). PRS periods are divided into three times,
namely the PRS-1, PRS-2 and PRS-3. During the PRS-1, after receiving Study Results
Card (SRC/KHS), and get permission to register from faculty trustee, students can en-
roll subjects that have been opened by the department. For example, the list of courses
that are opened by PCU Informatics department could be seen in Figure 4.
 Fig. 4. The list of courses that are opened for one semester

Then in Figure 5 you could see the example of courses that have been taken by students
that called the transcript.

Fig. 5. Student transcript example

Based on the example of courses offered (Figure 4) and student transcript (Figure 5),
the list of courses that can be taken by the student in the next semester could be seen in
Table 2.

Table 2. The example of courses which could be taken by student base on Figure 4 and 5

Course Code and Name Class Type Prerequisite

TF4251 – Komunikasi Data A,B,C Mandatory K. RD
TF4219 – Struktur Data A Mandatory K. PBO
TF4229 – Basis Data A Mandatory -
 For courses with prerequisites "Had attended", if the student has taken the prerequi-
site courses but did not passed, he could take courses with these prerequisites. But for
courses with prerequisites "Should be passed", the courses prerequisites should be
passed before the desired courses taken. Courses who had been taken by a student, still
could be taken again even after passed, with the aim to improve the credit point.
For courses with 'Elective' type, it can be taken if it meets prerequisites and the min-
imum number of credits that have been completed by the student is 60 credits. For
courses with 'Mandatory' type, it could be taken when the student have met the prereq-
uisites of the courses.
When registering on the PRS, students are given choice of priority 1 and 2. For ex-
ample, if a student has a value of 20 for priority 1, then the student should only enroll
20 Semester Credit Units (SCU/SKS) with priority 1, and the others are priority 2. The
priority 1 is very important because the chance it is likely to be accepted in the class
larger than the one with priority 2.
During the PRS-2 students could revise or add more courses to be taken up to a
maximum limit of credits. In addition, students could cancel courses which has received
in PRS-1. While the PRS-3 is the time to cancel courses that have been received in the
PRS-1 and PRS-2.

4 Genetic Algorithm Design

4.1 Chromosome Design

Chromosome is an important factor in the genetic algorithm, in which the design of
chromosomes will affect how the crossover implemented. Form of chromosome could
be seen in Figure 6.

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Day 0 [] [] [] [] [] [] [] [] [] [] [] [] [] [] []
Day 1 [] [] [] [] [] [] [] [] [] [] [] [] [] [] []
Day 2 [] [] [] [] [] [] [] [] [] [] [] [] [] [] []
Day 3 [] [] [] [] [] [] [] [] [] [] [] [] [] [] []
Day 4 [] [] [] [] [] [] [] [] [] [] [] [] [] [] []
Day 5 [] [] [] [] [] [] [] [] [] [] [] [] [] [] []
Fig. 6. Chromosome Design

The chromosome has day index starting from zero to the fifth indexes. Where in each
day they have hour index starting from zero to the fourteen indexes.
Day index only have six days, because on the day of Sunday is certainly no lectures.
As for the hour index in each day, containing fourteen ranging from 07.30 to 22.30 with
a time range of 1 hour. For example of chromosomes that have been filled could be
seen in Figure 7.
 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Day 0 [2 ] [2 ] [2 ] [0] [0] [6 ] [6 ] [0] [0] [0] [0] [0] [0] [0] [0]
Day 1 [0] [0] [49] [49] [0] [0] [0] [0] [10] [10] [0] [0] [0] [0] [0]
Day 2 [0] [0] [0] [0] [0] [0] [0] [0] [0] [0] [0] [0] [0] [0] [0]
Day 3 [0] [0] [0] [0] [0] [0] [19] [19] [19] [0] [0] [0] [0] [0] [0]
Day 4 [0] [33] [33] [33] [0] [0] [0] [0] [0] [0] [0] [0] [0] [0] [0]
Day 5 [0] [0] [0] [0] [44] [44] [44] [0] [0] [0] [0] [0] [0] [0] [0]

Fig. 7. Example of chromosomes that have been filled

The values present in a chromosome are the id of courses that opened in a semester.
Complete data from the ids are stored in a database table with the design as shown in
Table 3.

Table 3. Id Class information

ID class 49
Course name Struktur Data
Class A
SKSk 2
SKSp 1
SKSs 0
SKSr 0
SKS prerequisites 0
Course prerequisites Pass PBO, Pass AP
Course Start 08.30
Course Finish 10.30
Exam Schedule Day 3, 10.30
Specialization Course No

4.2 Crossover and Mutation Design

In this study, crossover design was made with the rule that a slot of a particular course
can only be subjected to a cross for all of the time slots of the course, and crossed to all
of the time slots of another course in the other chromosome. After that the crossover
results need to be checked whether valid or did not. If invalid, then the result was not
used in subsequent process. The example could be seen in Figure 8.
 Chromosome A
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Day 0 [2 ] [2 ] [2 ] [0] [6] [6 ] [0 ] [0] [0] [0] [0] [0] [0] [0] [0]

0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Day 0 [0] [0] [0] [7] [7] [7 ] [22] [22] [22] [0] [0] [0] [0] [0] [0]

Chromosome B

Fig. 8. The example of crossover

Whilst the mutation is done by randomly selecting two courses that have the same
length of time in the one of crossover result chromosome, the offspring, and then swap
their positions.

4.3 Fitness Cost

To determine the quality of the chromosome that is created, fitness cost calculation is
required. Here are the fitness cost calculation formulas of a chromosome. We adopt the
fitness cost calculation formulas from the similar formulas that was used by researchers
in the previous work [7].

 Fitness Cost based on Credit Point Units/SKS (FCSKS)

𝐹𝐶𝑆𝐾𝑆 = totalCreditPointUnit ∗ 30 (3)

 Fitness Cost based on the semester to be taken (FCSM)

𝐹𝐶𝑆𝑀 = ∑𝑚
𝑖=1 𝑐𝑜𝑢𝑟𝑠𝑒𝑆𝑒𝑚𝑒𝑠𝑡𝑒𝑟[𝑖] (4)

Where:
If the course is mandatory for next semester:
courseSemester = (courseSemester [i] - semesters) * (-5)
If the course is mandatory for the last semester:
courseSemester = (semesters - courseSemester [i]) * (200)
If the course is mandatory for the current:
courseSemester = 100
m = number of existing courses

 Fitness Cost based on curriculum (FCKUR)

 𝐹𝐶𝐾𝑈𝑅 = ∑𝑚
𝑖=1 𝑐𝑜𝑢𝑟𝑠𝑒C𝑢𝑟𝑟𝑖𝑐𝑢𝑙𝑢𝑚[𝑖] (5)

Where:
If the elective courses taken curriculum matches the user specialization:
courseCurriculum = 175
If the elective courses taken curriculum does not match the user specialization:
courseCurriculum = -75
m = number of courses in the chromosome

 Fitness Cost based on the interval between courses (FCJMK):

𝐹𝐶𝐽𝑀𝐾 = ∑𝑛𝑖=1 𝑑𝑎𝑦𝐼𝑛𝑡𝑒𝑟𝑣𝑎𝑙[𝑖] (6)

Where:
dayInterval = 300 - (interval between courses * 90)
n = number of day

 Fitness Cost based on courses to be taken (FCMKI)

𝐹𝐶𝑀𝐾𝐼 = ∑𝑛𝑖=1 ValueDesireCourse[𝑖] (7)

Where:
n = number of existing courses on chromosome
If the user wishes 'Want to Take':
ValueDesireCourse = 100
If the user wishes 'Extremely Want to Take':
ValueDesireCourse = 1000
If the user wishes 'Less Want to Take':
ValueDesireCourse = -200

 Fitness Cost based on the time desired by the user (FCWU)

𝐹𝐶𝑊𝑈 = ∑𝑛𝑖=1 𝑐𝑜𝑢𝑟𝑠𝑒𝐷𝑎𝑦Desired[𝑖] (8)

𝑐𝑜𝑢𝑟𝑠𝑒D𝑎𝑦Desired = ∑𝑘𝑗=1 𝑡𝑖𝑚𝑒𝑉𝑎𝑙𝑢𝑒[𝑗] (9)

Where:
n = number of day
k = 3, on the morning: k = 1, on the afternoon: k = 2, on the night: k = 3
If in accordance with the expected time:
timeValue = number of hours filled * 200
If not in accordance with the expected time:
timeValue = number of hours filled * -200

 Fitness Cost based on the maximum number of courses per day that is desired by the
user (FCMMK)
─ If user desirability ≥ number of courses,
 𝑛𝑢𝑚𝑏𝑒𝑟𝑂𝑓𝐶𝑜𝑢𝑟𝑠𝑒𝑠 [𝑖]
𝐹𝐶𝑀𝑀𝐾 = ∑𝑛𝑖=1 ∗ 300 (10)
𝑢𝑠𝑒𝑟𝐷𝑒𝑠𝑖𝑟𝑎𝑏𝑖𝑙𝑖𝑡𝑦[𝑖]

─ If user desirability ≥ number of courses

𝐹𝐶𝑀𝑀𝐾 = ∑𝑛𝑖=1 userDesirability[i] − 𝑛𝑢𝑚𝑏𝑒𝑟𝑂𝑓𝐶𝑜𝑢𝑟𝑠𝑒𝑠[i] ∗ 500 (11)

Where:
n = number of day

 Furthermore, the formula to calculate overall fitness cost of a chromosomme is as

follows:

Fitness Cost = FCSKS * w1 + FCSM * w2 + FCKUR * w3 + FCJMK * w4 +

FCMKI * w5 + FCWU * w6 + FCMMK * w7 (12)

Where:
w1 to w7 = weight of each fitness cost, value: 0...1.

5 Implementation and experiment results

System workflow could be seen in Figure 9.

 Insert
Semester and Start
Specialization

PRS-1 or PRS-2? PRS-1

PRS-2

Insert Transcript,
Day and Time
Is The PRS-1 No Preference, and
Schedule Exist?
desirability of the
courses.

Yes

Add and set Make courses

desirability of
schedules using
new courses or
delete existing Genetic
courses Algorithm

View or Print 5
End top Schedules to
user

Fig. 9. System Workflow

System starts with a login form to identify the user of the application. Furthermore, the
user can enter semester to be processed and areas of specialization are taken, as shown
in Figure 10.
 Fig. 10. Form Login and form setting semester and specialization

Later the user can choose to process the PRS-1 or PRS-2. Next will be displayed to the
user some forms in sequence, as shown in Figure 11 to 12. Here the user can enter the
latest transcript, day and time he desire and the desirability settings for all courses that
could be taken for the next semester.

Fig. 11. Form to insert the latest transcript

Fig. 12. Form to choose preferences days and hours of courses, and the number of courses/day

Fig. 13. Form to determine the desirability of all the courses that can be taken

Hereinafter the user can press a button to generate schedules using Genetic Algorithm.
Once the process is complete then to the user would be presented 5 best schedules.
These 5 best schedules are taken from 5 chromosomes with the highest fitness costs on
the last population. The scheduling results form and also the look to be printed (HTML
files) could be seen in Figure 14 and Figure 15. In addition, before starting the process
of scheduling, user can change the settings of variables required by the GA. Form to
change the settings can be seen in Figure 16.
Fig. 14. Form to view 5 best the results of automatic courses scheduling

Fig. 15. The display of HTML files to print the schedule

 Fig. 16. Form setting for Genetic Algorithm process

To test the results of the system, especially the results of the courses schedule that is
generated automatically, we use several kinds of tests. The first test is testing the effect
of the maximum population against the processing time, and the best fitness cost value.
For this test, setting the mutation probability is 0.3. The test results could be seen in
Table 4.

Table 4. Tests on the maximum population size

Running Times (millisecond)

Maximum Average of
Population Generate Initial Genetic Algorithm Fitness Costs
Chromosomes Processes
50 5962.8 3266.6 4302.4
150 15070.8 17938.4 4782.4
300 23283.2 32652.8 5297.4

From the test results it can be seen that the greater the maximum limit of the popu-
lation, the longer the process takes time, yet the average value of the best fitness costs
also increased. This is because more and more of the population there, will be the more
diverse opportunities to generate chromosomes, so that eventually the fitness cost val-
ues of the chromosomes will increase.
The second test is the test of the effects of increased mutation probability values
allowed on offspring chromosomes. For this test, the maximum population size be set
at 150 chromosome. The test results can be seen in Table 5.
 Table 5. Tests on the mutation probability values

Running Times (millisecond)

Mutation Average of
Probability Generate Initial Genetic Algorithm Fitness Costs
Chromosomes Processes
0.1 11614 11935 4532.8
0.4 14621.4 12650 4712
0.7 15942.2 19960.6 5176

From the test results it can be seen that the greater the probability of mutation, the
longer the time required for the process, but the average fitness value of a chromosome
best cost also increased. This is because the higher the chance of mutation, the more
diverse the resulting chromosome will be are greater, so that eventually the fitness cost
value of the best chromosome will increase.
In addition we also did some other tests, such as testing the maximum number of
chromosomes are copied by elitism method from the old population to the new popu-
lation. We also tested the maximum number of 'elitism counter', this counter to count
how many the sequences of populations will produce the same best fitness cost values,
before the program stop the process. Of all the tests performed at last we obtained the
best values of each setting. These values are used as the default values setting of the
system. The amount of the default settings and the testing results using default settings
can be seen in Table 6 and Table 7.

Table 6. Default setting

Setting Type Value

Maximum population 150
Max Elitism 15
Elitism Counter 15
Mutation Probability 0.7

Table 7. Testing results using default setting

Running Times (millisecond)

Number
Generate Initial Genetic Algorithm Scores
of Tests
Chromosomes Processes
1 23523 34653 5317
2 17116 18932 5317
3 17177 34653 5317
4 18592 32155 5269
5 33025 34090 5317
 Average: 21886.6 30896.6 5307.4

Of testing by using the default settings could be seen that the GA produces the best
fitness cost value greater compared to other settings that have been tested. Moreover, it
can be seen that in general the system will be convergence on a particular chromosome.
The final testing is testing in the form of questionnaires to potential users, in example
students majoring in Informatics department of Petra Christian University. The results
summary of these questionnaires are as follows: As many as 60% of respondents gave
scores of 4 and 40% gave scores of 5, of the scale from 1 (worst) to 5 (best), for ease
of use of this application. From these results it can be concluded that the user interface
of the application is easy to use. Meanwhile, for the question about the quality of the
generated schedules, 70% of respondents gave scores of 4 and 30% gave scores of 5,
of scale from 1 (worst) to 5 (best). Therefore, it can be concluded that in general the
respondents felt be helped by the application that could generate courses schedule au-
tomatically.

6 Conclusion

From the tests it can be concluded that the automatic scheduling system is well made.
Processing speed is also good, averaging less than 1 minute. From the test results on
the potential users can be concluded that the interface has a good design and user
friendly. In addition, the automatically generated class schedules are also correct and
in accordance with the expectations of users.

References
1. Mitchell, M.: An Introduction to Genetic Algorithms. United States of America: Massachu-
setts Institute of Technology. (1996).
2. Negnevitsky, M.: Artificial Intelligence: A Guide to Intelligent Systems 2nd Ed. Pearson
Education Limited. (2005)
3. Goldberg, D.E.: Genetic Algorithms in Search, Optimization and Machine Learning. Addi-
son-Wesley, Reading, MA. (1989).
4. Davis, L.: Handbook on Genetic Algorithms. Van Nostrand Reinhold, New York. (1991).
5. Haupt, R.L., Haupt, S.E.: Practical Genetic Algorithms 2nd Ed. United States of America:
John Wiley & Sons, Inc. (2004).
6. Holland, J.H.: Adaptation in Natural and Artificial Systems. University of Michigan Press,
Ann Arbor. (1975).
7. Budhi, G.S., Handojo, A., Soloment, B.: The Use of compact-Genetic Algorithm (cGA) for
Petra Christian University Classroom Scheduling Optimization. In: Proc. National Seminar
on Information Technology. (2009).