UNIVERSITY TIMETABLE SCHEDULING USING

METAHEURISTIC ADAPTIVE-ELITIST GENETIC ALGORITHM

A CS Thesis Project

Presented to the Faculty of

AMA COMPUTER COLLEGE

LAS PIÑAS

In Partial Fulfillment of the

Requirements for the Degree of

BACHELOR OF SCIENCE IN COMPUTER SCIENCE

Researcher:

Celiz, Edcel A.

April 2018
 DEDICATION

This research is dedicated to myself and fellow software developers.

To remind that the question is not “can I make this?” but “how long till I

make this?”

II
 ACKNOWLEDGEMENT

I would like to acknowledge and thank the following:

Our God Almighty, for making all of this happen and serving as an

inspiration towards the future.

My beloved parents, Maribel and Carlos Celiz, and brother,

Christopher Celiz, who have always supported me in all their capabilities

and served as my inspiration for my success in life.

My beloved partner, Araceli Liza, who have expressed her endless

support, patience and love to me all throughout the research.

AMA Computer College Las Piñas’ dean, Engr. Michelle Pullon, who

made my research possible and guided my towards its development.

AMA Computer College Las Piñas’ college faculty who have shown

their support and knowledge which made the research possible.

Research panelists for creating right judgment and imparting their

professional views on the research.

My friends and peers who have shown their support and help in

times of distress.

III
 TABLE OF CONTENTS

Approval Sheet I

Dedication II

Acknowledgement III

Abstract IV

Table of Contents V

CHAPTER I

THE PROBLEM AND ITS BACKGROUND

Introduction 1

Background of the Study 2

Statement of the Problem 4

Significance of the Study 5

Conceptual Framework 6

Scope and Limitation 16

Definition of Terms 27

CHAPTER II

REVIEW OF RELATED STUDIES AND LITERATURE

Foreign and Local Studies and Literature 29

CHAPTER III

RESEARCH DESIGN AND METHODOLOGY

Research Design Methodology 37

V
Method of Software Development

Method of Software Evaluation 38

CHAPTER IV

PRESENTATION, ANALYSIS AND INTERPRETATION OF DATA

CHAPTER V

SUMMARY, CONCLUSION AND RECOMMENDATION

Summary 49

Conclusion 50

Recommendation 51

Figures

Figure 1.0 Simplified IPO-Model of the Program 6

Figure 2.0 Basic Genetic Algorithm Process 7

Figure 3.0 Davis’ Order Crossover 15

Figure 4.0 Iterative Development Methodology 38

Figure 5.0 Evaluation Tree (Simplified Branch) 41

Tables

Table 1.0 Evaluation Matrix 8

Table 2.0 Tournament Selection Example 14

Table 3.0 Constraints 17

Table 4.1 Instructor CSV Columns 21

Table 4.2 Rooms CSV Column

VI
Table 4.3 Subjects CSV Column

Table 5.0 Minimum System Requirements 22

Table 6.0 Genetic Algorithm Settings 23

Table 7.0 Selected Evaluation Matrix 42

Table 7.1 Balanced Distribution Evaluation Matrix 45

Table 7.2 Subject Placement Evaluation Matrix 46

Table 7.3 Tight Constraints Evaluation Matrix 47

Table 8.0 Random Settings Limitation 43

Table 9.0 Random Settings

Table 10.1 Model 1 Summary 46

Table 10.2 Model 2 Summary 47

Table 10.3 Model 3 Summary 48

Bibliography

Appendices

Appendix A – Letters

Appendix B – Title Proposal

Appendix C – College Curriculums

Appendix D – Sample Output

Appendix E – Program Listings

Appendix F – Application Results

Appendix G – Gantt Chart

Appendix H – Curriculum Vitae

Appendix I – Final Permit

VII
 CHAPTER 1

THE PROBLEM AND ITS BACKGROUND

Introduction

Timetable scheduling is the process of creating timetables that fit

the constraint of the scenario. It is used in a magnitude of industry from

scheduling transportations up to creating complex schedule for highly

optimized automated factories. Majority of small scale scheduling are done

manually while larger operations require computer assisted scheduling.

Artificial intelligence is one of the rising computing solution due to

an increase in computing power. It can be applied to different types of

problems which can help optimize existing solutions or create never been

tried solutions due to multiple limitations. Artificial intelligence helps

create solutions for non-deterministic polynomial time problems which

businesses and organizations will always have.

Genetic algorithm is a metaheuristic that mimics the process of

natural selection. It can be performed in multiple different ways with

different types but it will all follow the same concept. This research aims

to create an artificial intelligence through the use of evolutionary

algorithm, specifically genetic algorithm combined with adaptive and elitist

traits that can generate a university schedule timetable with the goal of

1
generating a valid and as optimal as possible solution with certain

constraints.

Background of the Study

Scheduling classes in an institution is one of the tasks that can be

categorized under operations management. Operations management aims

to maximize efficiency in a certain area. Scheduling classes is often done

by humans which proves to be efficient but does not mean it is perfect.

Using computers to find the best solution for this problem using

conventional method is extremely inefficient. Computers were therefore

ignored for a time when solving such problem. However, rise in computing

power and usage has opened ways in solving scheduling problems.

Operations management is often considered as backbone of many

companies. Getting the most efficient work means higher profit. In an

educational institution scenario, getting the most efficient schedule does

not only help reduce expense but mainly to cater to students. Often, better

schedules give students and instructors better control on their time.

Creating schedules is done better by humans than conventional

computing techniques because humans have powerful brain to assess

constraints and combinations better. However, scheduling is known to be

non-deterministic polynomial time (NP) complete problem where in order

to find the best solution, every possible combinations should be executed.

Humans cannot compute every possible combination and therefore solves

2
the problem by just filling up and making sure that the constraints are

met. This is known to be “good enough” solution. This process is prone to

errors, inefficiency and violation of constraints especially when working in

a highly tight scenario. Replicating the human way of solving the problem

(Greedy algorithm) in computers is extremely inefficient due to its

computing cost.

Computers are rarely used for solving this problem because of the

stigma that it is unreliable and inefficient. However, this is back when

computers are slow and not intended to solve compute intensive problems.

With the explosion of big data and exponential increase in computing

power, using computers to solve new sets problems that previously can

only be solved by humans became easy. Scheduling is one of the problems

that can now be solved by computers with solutions acceptable or even

better. Artificial intelligence has become more common today brought by

the growth in computing industry. Several solutions fall under artificial

intelligence such as machine learning, deep learning and genetic

algorithms. All of these processes have their own pros and cons. Genetic

algorithm is fit for scheduling problem as it creates a solution over time

based on rules and criteria.

In AMA Computer College Las Piñas, the college dean is responsible

for creating schedules. The scheduling is done semi-manually with

computer assistance. Multiple problems occur from this process. The

usability and friendliness of the software that assists in scheduling is a

3
burden due to a number of features that are scattered and not even being

utilized majority of the time. The process can still not provide a solution

that is ensured to be free from errors, bad looking result and constraint

violations. Though it has become a practice for the dean to make

schedules, problems in making scheduling solution is not gone. Since the

approval of K-12 education system in the Philippines, the institute has

admitted senior high school students which put the dean in creating a

schedule in a very tight condition. With the current conditions, using a

computer to help in solving the scheduling problem can be an option. This

research aims to create a candidate software solution by using genetic

algorithms to generate schedule.

Statement of the Problem

Scheduling in AMA Computer College Las Piñas is being done semi-

manually with computer assistance which can produce errors, violations

of constraints and inefficiencies noting that scheduling has gone harder

with tighter constraint. The problem can be broken down into;

 Scheduling is still being done semi-manually with the operator

using the process alike to greedy algorithm which can produce

results with errors.

 The software assisting the operator is not user friendly and the

experience of using the software is bad and far from standards. This

leads to confusion and inability to exert the software's full potential.

4
  The software being used is not fitted for the institute and therefore

creates a steep learning curve. The universality of the application

causes confusion and inefficiencies with preparation of scenario for

generation of solution.

Significance of the Study

For AMA Computer College Las Piñas: The system will greatly

reduce time and effort spent on generating a schedule, therefore giving

more time for the administrator to manage the institute. This will also be

applicable with the senior high school department of the said institute.

For AMA Education System and Other Educational Institutes:

Multiple educational institutes can adopt the system and benefit by having

a computed scheduling solution. It will help ease the management of

school and possibly reduce expenses by having a more utilized schedule.

The system is built to support scheduling for levels lower than college.

For the Researcher: Developing the system will greatly help the

researcher towards improving on computer science due to implementation

of multiple algorithms. It will also help provide a step for the growing

artificial intelligence community. The research is going to be included in

the portfolio which can display the skills of the researcher in tackling

complex problems.

5
 For Future Researchers: Researchers can use this research as a

source for their research. They may use this approach on solving

optimization problems or even create a more efficient version. This

research comes with helpful data on understanding the performance and

process which will help future researchers.

Conceptual Framework

The program can easily be described as a scheduling computing

software wherein you input basic data sets and it will output a structured

result. A simple input-process-model would be:

Input Process Output

 Instructors
Application of genetic
 Rooms
algorithm to generate result.
 Sections Class schedules
 Subjects
 Settings

Feedback

Figure 1.0 Simplified IPO-Model of the Program

Genetic algorithm is a computing method that natural selection

which is a process in biological evolution. The process is repetitive and

best conceptualized using a flowchart. In genetic algorithm, a solution is

referred as chromosome instead of individual. The figure below shows the

basic process of genetic algorithm. However, in practice of the system,

6
another step is added after evaluation which is adjustment of

environment.

Figure 2.0 Basic Genetic Algorithm Process

Generation of population is to be done randomly with a mix of

greedy approach where a random point will be selected and act as starting

point and then start filling up the table with any entity that would fit.

During generation of population, all hard constraints are to be met.

Medium constraints will undergo multiple attempts to be followed while

soft constraints are totally ignored.

Evaluation is the process of calculating the fitness of a

chromosome. Fitness is the measure of quality of the chromosome relative

to the desired solution. It also triggers the ending process given that a

certain fitness is met. Calculating the fitness is done by assessing each

chromosome to the selected medium and soft constraints specifically the

7
subject placement, lunch break, section rest, section idle time, instructor

rest, instructor load balance and meeting pattern.

The fitness calculation would be largely based on the amount of

plotted subject compared to the required subjects. However, the

calculation would still depend on the evaluation matrix provided by the

user. The evaluation matrix is a set of constraint weights that has the

capability to shape the solutions. It is a list of containing the prioritization

of constraints using a distribution of a hundred percent (100%)

Table 1.0 Evaluation Matrix

Name Function

Attempts to create valid schedule for all

Subject Placement
subjects of all sections.

Attempts to at least have a 30-minute break

Lunch Break
between 11:00 AM and 1:00 PM

Attempts to at least give sections a 30-

Section Rest minute break for every 3 consecutive hours

of session

Attempts to ensure that sections will have

little or no idle time with respect to lunch

Section Idle Time break and section rest*.

*Section rest is modified by removing the need

for consecutive session.

8
 Attempts to at least give instructors a 30-

Instructor Rest minute break for every 3 consecutive hours

of session

Attempts to balance the distribution of

Instructor Load
instructor load with respect to their pooled
Balance
subjects.

Attempts to ensure that all divided schedules

Meeting Pattern are following a pattern instead of random

date selection.

Computing the total fitness would use the formula below where:

= 𝑐ℎ𝑟𝑜𝑚𝑜𝑠𝑜𝑚𝑒, 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛𝑠 𝑓𝑜𝑙𝑙𝑜𝑤 𝑡ℎ𝑒 𝑒𝑣𝑎𝑙𝑢𝑎𝑡𝑖𝑜𝑛 𝑚𝑎𝑡𝑟𝑖𝑥 𝑏𝑒𝑙𝑜𝑤 𝑎𝑛𝑑 𝑓𝑊 𝑖𝑠 𝑒𝑣𝑎𝑙𝑢𝑎𝑡𝑖𝑜𝑛 𝑤𝑒𝑖𝑔ℎ𝑡

𝒇𝒊𝒕𝒏𝒆𝒔𝒔(𝒙) = (𝑔(𝑥) ∗ 𝑔𝑊) + (ℎ(𝑥) ∗ ℎ𝑊) + (𝑖(𝑥) ∗ 𝑖𝑊) + (𝑗(𝑥) ∗ 𝑗𝑊) + (𝑘(𝑥) 𝑘𝑊)

+ (𝑙(𝑥) ∗ 𝑙𝑊) + (𝑚(𝑥) ∗ 𝑚𝑊)

Subject Placement

This evaluation stage calculates fitness of the chromosome base on

the subject placement. Computing this would use the formula below

where:

𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑝𝑙𝑎𝑐𝑒𝑑 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑠

𝒇𝒊𝒕𝒏𝒆𝒔𝒔 = ∗ 100
𝑡𝑜𝑡𝑎𝑙 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑐𝑜𝑚𝑚𝑖𝑡𝑡𝑒𝑑 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑠

9
Lunch Break

This evaluation stage checks if sections has lunch break all

throughout the week. Attending section days are evaluated and tagged

“without lunch break” if all timeslot between 11:00 AM to 1:00 PM is filled.

Calculating the fitness follows the formula below:

𝑑𝑎𝑦𝑠 𝑤𝑖𝑡ℎ𝑜𝑢𝑡 𝑙𝑢𝑛𝑐ℎ 𝑏𝑟𝑒𝑎𝑘

𝒇𝒊𝒕𝒏𝒆𝒔𝒔 = ∗ 100
𝑡𝑜𝑡𝑎𝑙 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑎𝑡𝑡𝑒𝑛𝑑𝑖𝑛𝑔 𝑠𝑒𝑐𝑡𝑖𝑜𝑛 𝑑𝑎𝑦𝑠

Section Rest

This evaluation stage calculates the rest period of sections which is

dictated by the rule that there should be at least a break for every three

(3) consecutive hours of session. Calculating the fitness follows the

formula below:

𝑑𝑎𝑦𝑠 𝑤𝑖𝑡ℎ𝑜𝑢𝑡 𝑠𝑒𝑐𝑡𝑖𝑜𝑛 𝑟𝑒𝑠𝑡

𝒇𝒊𝒕𝒏𝒆𝒔𝒔 = ∗ 100
𝑡𝑜𝑡𝑎𝑙 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑎𝑡𝑡𝑒𝑛𝑑𝑖𝑛𝑔 𝑠𝑒𝑐𝑡𝑖𝑜𝑛 𝑑𝑎𝑦𝑠

Section Idle Time

This evaluation stage checks for the unnecessary gap between

subject to subject of sections. The section idle time makes use of modified

section rest evaluation. The requirement for consecutive session is

removed, therefore gives a section free 30 minutes of rest time for every

three hours of session present in the day. This stage also respects lunch

10
break and section rest. Therefore, idle time is computed using the formula

below:

𝒊𝒅𝒍𝒆 𝒕𝒊𝒎𝒆

= 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑣𝑎𝑐𝑎𝑛𝑡 𝑡𝑖𝑚𝑒𝑠𝑙𝑜𝑡𝑠 𝑏𝑒𝑡𝑤𝑒𝑒𝑛 𝑓𝑖𝑟𝑠𝑡 𝑎𝑛𝑑 𝑙𝑎𝑠𝑡 𝑠𝑒𝑠𝑠𝑖𝑜𝑛 𝑜𝑓 𝑡ℎ𝑒 𝑑𝑎𝑦

− 𝑎𝑙𝑙𝑜𝑤𝑒𝑑 𝑏𝑟𝑒𝑎𝑘𝑠

If idle time is more than zero for the day, then it should be tagged

with excess rest. We can therefore compute the fitness of section idle time

by:

𝑑𝑎𝑦𝑠 𝑤𝑖𝑡ℎ𝑜𝑢𝑡 𝑒𝑥𝑐𝑒𝑠𝑠 𝑟𝑒𝑠𝑡

𝒇𝒊𝒕𝒏𝒆𝒔𝒔 = ∗ 100
𝑡𝑜𝑡𝑎𝑙 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑎𝑡𝑡𝑒𝑛𝑑𝑖𝑛𝑔 𝑠𝑒𝑐𝑡𝑖𝑜𝑛 𝑑𝑎𝑦𝑠

Instructor Rest

This evaluation stage calculates the rest period of instructors which

is dictated by the rule that there should be at least a break for every three

(3) consecutive hours of session. Calculating the fitness follows the

formula below:

𝑖𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑜𝑟 𝑑𝑎𝑦𝑠 𝑤𝑖𝑡ℎ𝑜𝑢𝑡 𝑟𝑒𝑠𝑡

𝒇𝒊𝒕𝒏𝒆𝒔𝒔 = ∗ 100
𝑡𝑜𝑡𝑎𝑙 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑡𝑒𝑎𝑐ℎ𝑖𝑛𝑔 𝑖𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑜𝑟 𝑑𝑎𝑦𝑠

Instructor Load Balance

This evaluation stage computes the normalized distribution of

instructor load. This takes into account the fact that instructors may have

a mutual subject pool with other instructors. The following formula below

shows the formula for calculating the load balance:

11
𝒊𝒏𝒔𝒕𝒓𝒖𝒄𝒕𝒐𝒓 𝒍𝒐𝒂𝒅(𝒊𝒏𝒔𝒕𝒓𝒖𝒄𝒕𝒐𝒓)

𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑙𝑜𝑎𝑑 𝑡𝑎𝑘𝑒𝑛 𝑏𝑦 𝑖𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑜𝑟

=
𝑡𝑜𝑡𝑎𝑙 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑙𝑜𝑎𝑑 𝑖𝑛 𝑡ℎ𝑒 𝑠𝑢𝑏𝑗𝑒𝑐𝑡 𝑝𝑜𝑜𝑙 𝑜𝑓 𝑖𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑜𝑟

∑ 𝑖𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑜𝑟 𝑙𝑜𝑎𝑑(𝑖𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑜𝑟𝑠)
𝒇𝒊𝒕𝒏𝒆𝒔𝒔 =
𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑎𝑐𝑡𝑖𝑣𝑒 𝑖𝑛𝑠𝑡𝑟𝑢𝑐𝑡𝑜𝑟𝑠

Meeting Pattern

This evaluation stage checks for the correction in the usage of

meeting patterns. There are only currently two (2) defined meeting patterns

which is “MWF” and “TTH”. Calculating the fitness follows the formula

below:

𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑠 𝑎𝑏𝑖𝑑𝑖𝑛𝑔 𝑚𝑒𝑒𝑡𝑖𝑛𝑔 𝑝𝑎𝑡𝑡𝑒𝑟𝑛

𝒇𝒊𝒕𝒏𝒆𝒔𝒔 = ∗ 100
𝑡𝑜𝑡𝑎𝑙 𝑎𝑚𝑜𝑢𝑛𝑡 𝑜𝑓 𝑐𝑜𝑚𝑚𝑖𝑡𝑡𝑒𝑑 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑠 − 𝑠𝑢𝑏𝑗𝑒𝑐𝑡𝑠 𝑖𝑛𝑑𝑖𝑣𝑖𝑠𝑖𝑏𝑙𝑒

After evaluation, an additional step is added in order to perform the

adaption of artificial intelligence to its current performance. Adaption is

the process of changing running variables to cater for better results. This

helps the artificial intelligence to avoid focusing on one set of problem

which may cause pre-mature convergence. Adaption performs two

important task which is population alignment and mutation rate

adjustment.

12
Population Alignment

Population is the term for the group of solutions. Before generation

of solution, the operator using the system can change the running

variables of the algorithm. The settings “Minimum Population” and

“Maximum Population” defines the limit of population change. Calculation

of population alignment will show how much change in population is to be

done using a series of formula below:

𝒄𝒉𝒂𝒏𝒈𝒆(𝒙)

∑ 𝑖𝑛𝑠𝑡𝑎𝑛𝑐𝑒𝑠 𝑜𝑓 𝑥 𝑤𝑖𝑡ℎ (−1 < 𝑑𝑒𝑣𝑖𝑎𝑡𝑖𝑜𝑛 < 1)

( ∗ 100) − 𝑙𝑜𝑤 𝑣𝑎𝑟𝑖𝑒𝑡𝑦 𝑠𝑒𝑡𝑡𝑖𝑛𝑔𝑠
𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑐𝑜𝑢𝑛𝑡
=
100

∗ 𝑝𝑜𝑝𝑢𝑙𝑎𝑡𝑖𝑜𝑛 𝑐𝑜𝑢𝑛𝑡

Mutation Rate Adjustment

Mutation rate is the chance for each chromosome to get random

change. Mutation rate adjustment happens when the trigger which is set

before running is met. When a generation completes without triggering the

adjustment, the mutation rate will decrement by 0.5% else it will be

increased by 0.5%. Calculation of mutation rate adjustment trigger follows

the formula below:

𝒄𝒉𝒂𝒏𝒈𝒆 = 0.5 ∗ (−1 𝑖𝑓 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑓𝑖𝑡𝑛𝑒𝑠𝑠 − 𝑝𝑟𝑒𝑣𝑖𝑜𝑢𝑠 𝑎𝑣𝑒𝑟𝑎𝑔𝑒 𝑓𝑖𝑡𝑛𝑒𝑠𝑠

< 𝑎𝑑𝑗𝑢𝑠𝑡𝑚𝑒𝑛𝑡 𝑡𝑟𝑖𝑔𝑔𝑒𝑟 𝑠𝑒𝑡𝑡𝑖𝑛𝑔𝑠 𝑒𝑙𝑠𝑒 1)

13
 After evaluation, chromosomes are picked to participate for

reproduction. Chromosomes with higher fitness is more likely to get

picked. This is where the survival of the fittest comes. There are multiple

ways of selecting chromosomes all of which have their pros and cons.

Having the most appropriate type of selection will help avoid early

convergence and promote diverse solutions. As an elitist variant of genetic

algorithm the top n% of population is guaranteed to proceed in the next

generation with same genes depending on the settings.

The rest of the selection process is to be done by implementing

multiple tournaments on the population. The participants in the

tournament are selected randomly with a quantity of 4% of existing

population but capped at twenty five (25). The tournament will run until

enough pairs are picked. The following example below shows a simulation

where 40% of the 10 chromosomes are selected in the tournament.

Table 2.0 Tournament Selection Example

Chromosomes Fitness Participant

#1 4.1 TRUE

#2 3.4

#3 5.6 TRUE

#4 7.5

#5 6.5

#6 1.2

#7 3.5 TRUE

#8 2.8 TRUE

14
 #9 7.5

#10 1.1

Winner: Chromosome #3

If there are 100 chromosomes in a population with a 10% elitism

rate, then we have to generate tournament 90 times in order to pair 45

chromosomes that will produce 90 offspring. The offspring and elite

chromosomes after crossover would total into 100 which is the population

count all throughout the process.

The mating pool consist of selected chromosomes that will undergo

crossover. The crossover operation involves production of offspring based

on genes of parents analogous to its biological counterpart. There are

multiple ways to perform crossover and like selection, it is important to

use the most appropriate one to have better outcome. Davis’ order

crossover works better for permutation based problems. The process

works by taking the gene cluster of first parent into the offspring and then

wrapping the gene cluster of second parent into the offspring. For each

pair of chromosomes, two offspring will be produced where parent’s role is

reversed to produce different offspring. All offspring will take their parent’s

place in the population. This can be visualized below where:

Figure 3.0 Davis’ Order Crossover

15
 Mutation is a genetic operator that helps maintain the variety of

solutions. It can also prevent solutions from being stuck in local optima.

Mutation is the process of altering the gene/s of a chromosome. Usually,

mutation rate is low and fixed. However, adaptive genetic algorithm means

that mutation rate may vary depending on the performance of the

population.

Once the genetic algorithm terminates either by force or meeting the

end criteria, the top 5 fittest chromosomes will be displayed for user

review. The user may pick any from the proposed solution and get its

output in a CSV file.

Scope and Limitation

This research aims to create an artificial intelligence that can create

timetable schedules. This only covers processing input data and

generating human sensible results. The process to be used in creating

schedules is adaptive-elitist genetic algorithm. A type of evolutionary

algorithm that involves keeping the best set of solutions over next

generations to preserve high level solution but at the same time, variables

such as the population count, mutation rate and selection pressure,

change in order to avoid premature convergence that leads to poor output.

16
 Computing the schedule would involve validation for each

constraint. Constraints are set of rules that directs the acceptance of the

output. There will be three types of constraints to be used;

1. Soft Constraints – A set of rules that can be broken without affecting

the validity of the output.

2. Medium Constraints – A set of rules that can be broken with an effect

to the validity of the output. However, this can only be broken if the

scenario is logically invalid or impossible.

3. Hard Constraints – A set of rules that would produce an invalid

solution if broken.

Table 3.0 Constraints

Hard Constraint Medium Constraint Soft Constraint

Students should

Sections’ subjects have only 30

Instructors teach one class
are placed on the minutes break for
at a time
schedule every two hours of

session per day

Sections should have Subjects that are

at least 30 minutes divided should

Instructors teach at their vacant time between follow the two

available schedule 11:00 AM to 1:00 PM defined meeting

for lunch break pattern which is

(Optional) "MWF" and "TTH"

17
 Instructors should
Sections should have
Instructors can only take have normalized
at least 30 minutes
N amount of subjects distributed load
vacant time for every
dependent to their based on the
3 consecutive hours
maximum amount of load instructor pool of
of session
subjects

Instructors should

have at least 30

Sections attend one class minutes vacant time

at a time for every 3

consecutive hours of

session

Sections attend at their

available schedule

Sections will stay at one

room unless taking a

laboratory subject

Sections with subjects that

are shared to other

sections should produce a

schedule that is

compatible to all sharing

participants

18
 Rooms can only take class

at its available schedule

Subjects have a right room

type

Subjects will divide

depending on its

configuration

Subjects will be assigned

an instructor if there is a

pool of instruction on its

configuration

The application takes five major input; Instructors, rooms and

subjects component supports addition of entry using the application and

importation. The sections component allows creation of sections with

special feature of sharing subject with other sections and the scenario

manager component handles running configuration. There would be two

types of output; using the result viewer of the application which can view

the top five solutions of the last generated scenario and exporting the

selected solution which will produce three comma separated values (CSV)

files. Tests for reliability and performance of the program specifically the

evaluation for the performance of the artificial intelligence will come from

usage of application using actual records of AMA Computer College Las

Piñas. The application will be using the Python language. The research will

19
be conducted during the second and third trimester of the school year

2017-2018 in AMA Computer College Las Piñas.

Results of the application is not guaranteed to be the best possible

solution for the scenario. The quality of the result relies heavily on running

preference. Due to its stochastic nature, results may vary from poor to

excellent. The system does not guarantee that every solution’s hard

constraint will be met especially when the presented scenario is logically

impossible or intensely tight. This also means that when in a strict

scenario soft or medium constraints are subject to violation in order to

meet hard constraints. Hard constraints are bound to be followed and will

never be violated. Importation of values from CSV supports strict

formatting shown below.

CSV Formatting

All CSV files should use line 1 of the file as file indicator. These

should be one of the following “instructors”, “rooms” and “subjects”. The

second line of the file should be used for defining table columns in same

order with the writing below:

20
1. Instructors – name, hours

Table 4.1 Instructors CSV Columns

Column Description Type (Options)

Name of the
name String
instructor

Serviceable hours
hours Integer (1-100)
per week

2. Rooms – name, type

Table 4.2 Rooms CSV Columns

Column Description Type

name Name of the room String

type Type of the room String (lec, lab)

3. Subjects – code, title, type, hours, splittable

Table 4.3 Subjects CSV Columns

Column Description Type

Short code of the

code String
subject
title Name of the subject Integer (1-100)

type Type of the subject String (any, lec, lab)

Hours per week of Double +.5 (1.0,

hours
the subject 12.0)

If the subject is going

splittable Boolean-Integer (0-1)
to be splitted

21
Example Importable CSV File

first_trimester_subjects.csv

subjects
code,title,type,hours,splittable
ENGL101,Communication Skills 1,lec,3,1
MATH101,College Algebra,lec,3,1
MATH110,College Trigonometry,lec,3,1
CS101,Introduction to Computing,lec,2,1
CS101L,Introduction to Computing,lab,2,0

The application offers a way to customize the running configuration

for genetic algorithm which can cause performance issues and improper

result if tweaked wrongly. Computers that will run the system should

follow the minimum required specifications shown below. The system is

not liable for any 3rd party damages that may occur due to inappropriate

use.

Table 5.0 Minimum System Requirements

Minimum System Requirements

Operating System: Windows 7 and Above

CPU: 2.0+ GHz with multithreading support

1 GB*
*This still relies on the scenario size and
RAM:
algorithm configuration see Chapter 5 for
assessment of consumption.
At least 1 GB Available Space*

*User generated files are excluded in assessment

Disk Space:
and minimum space is also subject to application

usage.

22
Genetic Algorithm Settings

The configuration of genetic algorithm within the application is not

one for all type where it can cater every scenario. This is due to the

limitation of computational power. To counter the limitation and still be

able to support scalability of usage, the algorithm’s running configuration

can be altered according to operators’ device capability or preference. The

following table below shows the description and recommendation for each

modifiable setting property of the algorithm.

Table 6.0 Genetic Algorithm Settings

Settings Description Limit

The minimum and starting number of

Minimum
chromosomes in a generation. Fifty (50)
Population 50-10,000
is the suggested value for this as it is
Count
high enough for a variety of solution.

The limit on how many chromosomes

can a generation have. Calculating how

much the computing device can handle

depends on its RAM. For every seventy

Population (70) active entities in a scenario (subject,

50-10,000
Count instructor, room, and section), it will

consume one megabyte (1 MB) of

memory per chromosome. Therefore, if

there are using fifty (50) entities and

maximum population count of one

23
 hundred (100), the peak consumption of

RAM would approximate to 72 MB plus

25% of the value for main application

handling. A low-end computer should

maintain low maximum population

count, it can match the minimum

population suggestion of fifty (50).

The limit of how many generations the

algorithm can produce. Upon reaching

the set value, the algorithm will stop.

The higher entity count or the more

Maximum complex constraints the scenario

50-10,000
Generations presents, the higher this should be

valued as it gives more chance for the

artificial intelligence to fix minor

problems or even create a breakthrough

that can change the entire solution set.

The maximum amount of attempts for

the artificial intelligence to place a valid

Maximum subject schedule in the chromosome. A

1,500-
Creation strong computer can put higher value on
30,000
Attempts this as this serves more as a brute force

tactic which can help increase the

quality of the solution.

24
 The threshold value for triggering

mutation rate adjustment. The value

checks for the difference of the latest

average fitness to its previous average

fitness. If the difference falls below or

Mutation Rate equals to the set value, the mutation

0.00-
Adjustment rate will increase by 5% while reducing it
100.00
Trigger by 5% if not. The value can be set to zero

if the operator does not want to change

the existing mutation rate of 10%. It is

recommended to keep the value of this

setting to its default of 0.08 to prevent

massive changes on the population.

The trigger for ending the algorithm.

Once a chromosome hits the set value,

the algorithm will stop. If a perfect

solution based on model is desired, this

Maximum can be set to 100%. However, the chance

0-100%
Fitness of getting a perfect solution on every

combination of model is not guaranteed.

It is suggested to place an optimal value

for this which can range from 90-98%

for acceptable level.

Elite The percent of how much of the current

0-100%
Population population’s best performing

25
 chromosomes will be preserved until the

next generation. The suggested value for

this is 5% as it keeps a lower count of

good chromosomes and helps prevent

premature convergence.

Measures the richness of the

population’s solution set. Each

chromosome’s fitness is measured and

normalized for statistical deviation. If

one of the deviation sigma goes beyond

the set value, the population count

changes by how much percent it is off

Deviation
else the population will be reduced by 0-100%
Tolerance
how much percent it lacks. This helps in

maintaining the variety of solutions

basing on the deviation of fitness

between chromosomes. The suggested

value is 55% as it prevents total

domination of a common group of

chromosomes.

The suggested values do not guarantee the best combination. The

computation of space consumption does not fully depict the actual usage.

26
Definition of Terms

Operational Terms:

Chromosome - A set of parameters which defines a proposed solution that

the genetic algorithm is trying to solve.

Crossover - A genetic operator used to vary the programming of a

chromosome or chromosomes from one generation to the next. It is

analogous to reproduction and biological crossover, upon which genetic

algorithms are based.

Deep Learning - A part of a broader family of machine learning methods

based on learning data representations, as opposed to task-specific

algorithms. Learning can be supervised, partially supervised or

unsupervised.

Gene - The basic unit of a chromosome. It represents as a part of a

solution.

Genetic Algorithm - A metaheuristic inspired by the process of natural

selection that belongs to the larger class of evolutionary algorithms (EA).

Genetic algorithms are commonly used to generate high-quality solutions

to optimization and search problems by relying on bio-inspired operators

such as mutation, crossover and selection.

Genotype - The genetic representation of the solution, in a basic solution,

it might be represented as a string of binary.

Machine Learning - A field of computer science that gives computers the

ability to learn without being explicitly programmed.

27
Mutation - A genetic operator used to maintain genetic diversity from one

generation of a population of genetic algorithm chromosomes to the next.

It is analogous to biological mutation.

NP-Complete - A complexity class where solutions can be verified quickly

and if there is a quick algorithm to solve this type of problem, it can be

used to solve other NP problems too.

Phenotype - The real world representation of the solution.

Population - A set of proposed solutions.

Selection – Stage of GA that selects chromosome for future combinations.

Technical Terms:

Chromosome - A DNA molecule with part or all of the genetic material of

an organism.

Crossover - A point or place of crossing from one side to the other.

Gene – Sequence of DNA which codes for a molecule that has a function.

Genotype – The part of the genetic makeup of a cell.

Mutation – The permanent alteration of nucleotide sequence of the

genome of an organism.

Phenotype – The composite of an organism’s observable characteristics or

traits.

Population – All the inhabitants of a particular location.

Selection – The action of carefully choosing someone as being the best.

28
 CHAPTER 2

REVIEW OF RELATED LITERATURE AND STUDIES

University timetable scheduling is a combinatorial problem where

finding the best solution is hard when the search space goes big. Knowing

the background of the problem helps understand why even at computing

era, solving it is a somewhat monumental task. Getting to know the

possible solutions for the problem is also needed as it might show some of

the shared traits that might lead to a breakthrough in finding a better

solution. Finally, gathering information regarding the proposed solution

will help guide the research into forming better solution.

1. Operations Research

The article published in Universal Class [1] and in the chapter 14 of

the book published in IBS center for management research [2] it is stated

that operations management is a delicate part of business which is

responsible for the process and quality of work. It is said that scheduling

plays a major role in meeting the customer’s requirements on time and

improving the efficiency of the overall system. In a report by R. Dan Reid

and Nada R. Sanders [3], they said that the theory of constraints is that the

result of a schedule is determined by the constraints. In order to avoid

getting overwhelmed by the constraints, one must identify bottlenecks in

the process and try to improve it.

29
2. Common Scheduling Conditions in the Philippines

In a university-based news article published by Althea Gonzales

and Jessy Go [4], students have expressed mixed reactions on their

university’s newly imposed scheduling rule. The classes were made earlier

and a break of fifteen (15) minutes were given between each subject. It is

said that it was implemented in order to give students time to prepare for

their classes and avoid missing out. Educational institutes sometimes

implement their own scheduling schemes in order to compensate to their

situations. In a recent independent news assessment of K-12’s status in

the Philippines written by Anne Marxze Umil [5], many institutes still lacks

in resources needed to accommodate students. There are still shortages in

classrooms and teachers. This also includes private schools who had to

receive a sudden influx of students supported by the voucher system. In

order to solve the problem on lack of resources, according to an article

written by Careen L. Malahay and Tweeny M. Malinao [6], schools have

implemented a shorter scheduling scheme to create shifts in order to

accommodate the students. The effects of shortages in resources can also

be felt in college level. In the article published by Paulyn Navarrete and

Illiana Tan, it has showed the dismay of students who cannot create a

comfortable schedule due to few subject openings and conflicts. This is

common among institutes who of course cannot handle every scenario.

30
3. How Humans Solve Complex Problems: The Case of the Knapsack

Problem

According to an article in University of Colorado [8], knapsack

problem is one of the classic problems in the field where computing

schedules belong. It is a combinatorial optimization problem with the goal

of finding the subset of items with the highest total value from a set of

items with given weight and value subject to weight constraint. In the

research about how humans solve complex problems written by Carsten

Murawski and Peter Bossaerts, it is found that humans solve the knapsack

problem by implementing a greedy algorithm process of thinking in the

early to middle phase of solving the problem followed by a branch and

bound like searching technique. In the experiment where the average

number of full knapsack configuration is one thousand three hundred

eighty one (1,381), the chance of getting the most optimal solution is

merely a 0.7%. The results of the study have shown that humans

performed extremely well with a median score of 97.4% of the most optimal

solution value. However, it is worth noting that just because humans can

solve knapsack problems with an optimal solution, it is not the same case

with scheduling problems.

4. NP-Complete Complexity Class

The knapsack problem belongs to the complexity class of non-

deterministic polynomial-time (NP) hard. These are problems where there

is no efficient solution algorithms yet. The computational time only grows

as the size grows according to Stephen A. Cook [10]. According to standard

31
defining organization’s computer scientist, Paul E. Black [11], NP-Complete

is a complexity class where solutions can be verified quickly and if there

is a quick algorithm to solve this class of problem, it can be used to solve

other NP problems too. In an article by J.D. Ullman [12], he said that finding

the most optimal schedule or solving scheduling itself belongs to NP-

Complete problem.

5. Ways on Solving University Class Scheduling

University class scheduling problem belongs to optimization

problems in operational research. It is considered as one of the most

difficult problems faced by universities and colleges today according to a

study on subject matter by Ahmed Wasfy and Fadi A. Aloul [13]. There are

multiple approaches towards solving scheduling problems according to

Ellis Cohen and Jarurat Ousingsawat [14]. Some are only feasible up to

certain level such as the constraints they can handle, objectives to meet

and computation time. The most common approaches are construction

algorithms which greedy algorithm is included and search algorithms

where tabu search, simulated annealing, particle swarm optimization,

genetic algorithm and many more from J Zhou, P E D Love, X Wang, K L

Teo and Z Irani’s review on scheduling [15]. Construction algorithms can

and sometimes mixed with search algorithms such as greedy and genetic

algorithm where the population may depend on the generated individuals

by the greedy algorithm.

32
6. Genetic Algorithms: An Overview

According to Melanie Mitchell’s lesson [16], genetic algorithm is a

program that mimics the process of biological evolution in order to solve

problems and to model evolutionary systems. The vast search space has

inspired the creation of genetic algorithm. Biological entities over the years

have adapted to its environment to be more efficient and effective. It has

then become part of the early waves of artificial intelligence wherein rules

and behavior are manually encoded. The algorithm is still in use in modern

period mostly for optimization problems.

7. Phenotypes, Genotypes, and Operators in Evolutionary Computing

In David B. Fogel’s publication [17], genetic algorithm works by

evolving a set of solutions into better ones. Each solutions are often

referred as chromosome. A group of chromosomes creates a population. A

solution is represented into two types, genotype and phenotype. The

genotype is the genetic representation of the solution, in a basic solution,

it might be represented as a string of binary. The phenotype is the real

world representation of the solution. In evolutionary computation, it is

important to convert phenotype into genotype as it is easier to compute.

Having a good and proper conversion will definitely help in computing the

solution.

8. Selection Methods for Genetic Algorithms

The process of genetic algorithm starts with initialization of

population. After initialization, evaluation of chromosomes are done to give

33
score on how did each solution perform. In order to continue into next

generation of solutions, selection operation is done. Selection is the

process of selecting chromosomes to be used as parents to produce next

generation of solutions. There are multiple types of selection which can be

used to support multiple types of problem. Some of these are roulette

wheel selection, stochastic universal sampling, linear rank selection,

exponential rank selection, tournament selection and many more

according to Khalid Jebari [18]. Tournament selection is done by selecting

random chromosomes wherein the best performing will be selected. This

will be done until enough chromosomes are selected. The performance of

tournament selection is incredibly fast but depends on its configuration.

It can either lead to fast good or bad convergence. However, the selection

pressure is high due to the amount of repetitive computation.

9. Crossover Operators in Genetic Algorithms: A Review

Crossover is a genetic operator which plays one of the biggest role

in genetic algorithm. It is done by creating an offspring based on the genes

of the parents mimicking the biological reproduction. There are many ways

on how to perform crossover but just like selection, each of them has their

own pros and cons. Some of these crossover types are standard crossover

using N-points, shuffle, discrete, uniform, order based and many more

according to A.J. Umbarkar and P.D. Sheth [19]. Order based crossover is

most appropriate for problems where groups or patterns are important. It

helps preserve existing combination while improving trying to improve it.

34
10. Genetic Algorithm Essentials, Studies in Computational Intelligence

The final operator in genetic algorithm is mutation. Mutation is a

random based operator that changes some gene in a chromosome. It helps

prevent early convergence by introducing random changes that might

increase or decrease the fitness of a chromosome. There are three

requirements in mutation; reachability, unbiasedness and scalability

according to O. Kramer’s study [20]. Usually, a fixed value for mutation rate

is used but adaptive variety of genetic algorithm may change the value

from generation to generation depending on the situation.

Scheduling is one of the most important part of a business. It plays

a big role in operations management as it helps maximize the efficiency of

production or service. In educational institutes, scheduling is one of the

biggest challenge often because bad schedules leads to students and other

party’s dismay. Solving the problem using only pure man work maybe

efficient enough but it will not scale up. To solve this problem, computing

the schedule using the right algorithm and constraints is a must. Genetic

algorithm presents itself as a valuable search optimization solution with

its capability to handle and manipulate data.

11. User Experience for Windows Desktop Applications

Microsoft’s Windows Development Center has created design

principles [20] for Windows desktop applications. Some of the important

considerations are categorized below:

35
 Windows – The application should be able to support 800x600

pixels resolution. It should be displayed initially on the center of

the user’s window.

 Layout – The size of controls and panes within a window match

their typical content.

 Text – Use ordinary, conversational terms and use title-style

capitalization for titles while sentence-case for all other UI

elements.

 Controls – Add label to controls that requires user input. Use

tooltips for unfamiliar features.

 Accessibility – The operator should be able to navigate the full

window using tab key.

 Dialogs – The dialog should present important data and

interaction.

36
 CHAPTER 3

RESEARCH DESIGN AND METHODOLOGY

Artificial intelligence is a computer science field that is being used

mostly on problems that includes precision or optimization. The research

falls under optimization problem and evaluation of performance in an

extensive verification is not a feasible solution. It is important to select an

appropriate research design and methodology to get the viability of the

research.

Research Design Methodology

The selected method for collecting and analyzing data for testing the

system is quantitative. The data to be collected are sets of school

configuration (e.g. subjects, instructors and rooms). These datasets will be

profiled after being used in the system. The profiling results shall

determine the performance of the system by performing artificial

intelligence systems based evaluation.

Method of Software Development

Iterative development as a method for creating the software fits for

continuously changing environment for artificial intelligence development.

This enables the developer to learn every iteration and apply it to future

instance. This methodology is fit for the research as the system relies on

37
tweaking and managing constraints. Each iteration is an enhancement to

the system which will repeat until the system delivers its expected result.

Figure 4.0 Iterative Development Methodology

Initial planning has been done by designing system architecture,

data schematic and models. Upon entering an iteration, the feature to be

done is decided and planned. There are two types of iterations used;

adding a feature or a model and tweaking values. Each iteration’s goal is

then implemented and then manually tested.

Method of Software Evaluation

The selected evaluation methodology for the research is a

combination of Monte Carlo methods and surrogate modelling. It is an

experimental methodology as evaluation for evolutionary computation

based scheduling systems are yet to be done.

The Monte Carlo methods are a class of computational algorithms

that uses repeated random sampling to obtain numerical results. It is often

used under the optimization and probability problem sets. It works by

tweaking the initial value and trying to produce a solution from it.

38
 The surrogate modelling is often used in science and engineering

field where the outcome cannot be easily measured directly so a model is

used instead. An example of this is finding the optimal design for a car’s

aerodynamics. Computing the design’s performance with real life like

simulations is expensive so instead, an emulator model is used. This

generally works by creating an approximation of models which mimics the

behavior of real model as closely as possible without the need for a more

expensive evaluation.

Implementation of Monte Carlo method and surrogate modelling for

evaluation of artificial intelligence based scheduling system, specifically

the researcher’s study shall follow the following steps:

1. Generation of a set of solutions using Monte Carlo method with genetic

algorithm settings as variables that can be tweaked.

2. Creation of a surrogate model which only uses the evaluation matrix

as basis.

3. Computation of each set of solutions generated by the Monte Carlo

method difference to surrogate model.

4. Providing the mean of the result as basis for performance of the

artificial intelligence (Scenario based).

Usage of Monte Carlo method in system is for generation of random

genetic algorithm configuration. The surrogate modelling usage for this

research makes use of an imaginary 1-dimensional metric basing on the

39
evaluation matrix. The model will serve as a basis for evaluating the

closeness of the solution to the ideal one. In conjunction with software

evaluation methodology using the proposed steps above, data gathering

would then lean towards data-centric approach. This is to provide correct

conclusions towards the conclusion of the research. The proposed data to

be gathered are the ones to be used by the evaluation above which will be

composed of mostly result performance grade (fitness).

40
 CHAPTER 4

PRESENTATION, ANALYSIS AND INTERPRETATION OF DATA

Using the surrogate modelling and combining it with Monte Carlo

method has presented multiple data for interpretation. The following figure

below shows the chain of data collection and evaluation methodology for

one model:

Figure 5.0 Evaluation Tree (Simplified Branch)

There will be three (3) models to be used listed on the table below.

For each model, there will be three (3) scenarios. For each scenario, there

will be three (3) randomly generated value of settings. It is important to

note that there will only be one time generation of random values for

settings and the three (3) generated values will be used all throughout the

scenarios to prevent bias in result. For every setting, there will be five (5)

41
generated result set. In total, there will be 135 solution set for evaluation

which will be abstracted per level.

Table 7.0 Selected Evaluation Matrix

Name Matrix
Subject Placement 16%

Lunch Break 14%

Section Rest 14%

Balanced Distribution Section Idle Time 14%

Instructor Rest 14%

Instructor Load Balance 14%

Meeting Pattern 14%

Subject Placement 100%

Lunch Break 0%

Section Rest 0%

Subject Placement Section Idle Time 0%

Instructor Rest 0%

Instructor Load Balance 0%

Meeting Pattern 0%

Subject Placement 20%

Lunch Break 0%

Section Rest 20%

Tight Constraints Section Idle Time 20%

Instructor Rest 20%

Instructor Load Balance 20%

Meeting Pattern 0%

42
 The randomization has a set amount of possible values based on

the computer’s optimal capability and suggested values from genetic

algorithm settings (Table 6.0 Genetic Algorithm Settings) which can be

viewed in the table below:

Table 8.0 Random Settings Limitation

Name Limit

Minimum Population Count 50-200

Maximum Population Count Minimum Population Count-200

Maximum Generations 50-150

Maximum Creation Attempts 1,500-4,500

0.00-1.00 where value is divided

Mutation Rate Adjustment Trigger
by five (5)

Maximum Fitness 90-100

Elite Population 0-10%

Deviation Tolerance 50-75%

Using randomization provided above, here are the three sets of

genetic algorithm settings:

Table 9.0 Random Settings

Entry Matrix

Minimum Mutation Rate Adjustment

63 .09
Population Count Trigger
#1
Maximum
84 Maximum Fitness 96%
Population Count

43
 Maximum
64 Elite Population 6%
Generations

Maximum

Creation 1,873 Deviation Tolerance 57%

Attempts

Minimum
Mutation Rate Adjustment
Population 79 .05
Trigger
Count

Maximum

Population 148 Maximum Fitness 93%

#2 Count

Maximum
89 Elite Population 7%
Generations

Maximum

Creation 2,279 Deviation Tolerance 69%

Attempts

Minimum Mutation Rate Adjustment

78 .11
Population Count Trigger

Maximum
96 Maximum Fitness 95%
Population Count

#3 Maximum
63 Elite Population 2%
Generations

Maximum

Creation 1,513 Deviation Tolerance 63%

Attempts

44
 There are three (3) scenarios prepared for the evaluation. The three

scenarios are imitation of the curriculum for each trimesters on AMA

Computer College Las Piñas have (Appendix C) for CS, IT and CpE courses.

Each scenario have 91, 93 and 95 entities respectively. These entities are

active instructors, rooms, subjects and sections. Each model below shows

the summarized version of each scenario. The summarization includes

computation of min, max and average of total fitness of each result from

all settings.

Model 1: Balanced Distribution

Table 7.1 Balanced Distribution

Name Matrix

Subject Placement 16%

Lunch Break 14%

Section Rest 14%

Section Idle Time 14%

Instructor Rest 14%

Instructor Load Balance 14%

Meeting Pattern 14%

Using the current model, no scenario solutions was able to reach

the maximum possible solution. The distribution of priority has made the

artificial intelligence have hard time balancing all the constraint. The table

45
below shows the performance of the algorithm to each scenario in

combined settings evaluation.

Table 10.1 Model 1 Summary

Scenario Highest Average Lowest Average

Fitness Fitness Fitness Time

1 84.45 80.32 78.61 21:28

(Settings 2)

2 84.50 81.01 78.94 23:30

(Settings 2)

3 84.48 80.67 79.0 22:48

(Settings 2)

Model 2: Subject Placement

Table 7.2 Subject Placement Evaluation Matrix

Name Matrix

Subject Placement 100%

Lunch Break 0%

Section Rest 0%

Section Idle Time 0%

Instructor Rest 0%

Instructor Load Balance 0%

Meeting Pattern 0%

Being one of the easiest model, the artificial intelligence was able to

adapt easily as it does not have to worry about other constraints when

46
placing entities. It has showed its superiority on initializing the population

to provide acceptable solutions even though it is just starting.

Table 10.2 Model 2 Summary

Scenario Highest Average Lowest Average

Fitness Fitness Fitness Time

1 100.00 99.69 97.15 4:29

(Settings 1)

2 100.00 99.16 93.46 5:19

(Settings 1)

3 100.00 99.69 97.15 5:16

(Settings 1)

Model 3: Tight Constraints

Table 7.3 Tight Constraints Evaluation Matrix

Name Matrix

Subject Placement 20%

Lunch Break 0%

Section Rest 20%

Section Idle Time 20%

Instructor Rest 20%

Instructor Load Balance 20%

Meeting Pattern 0%

The third model which can be considered as closest to real schedule

in terms of software capability was able to assess solutions that are much

47
better than the first model despite having the same nature of balance in

constraint prioritization. This has shown the system was capable of

creating solutions that does not violate too much constraint.

Table 10.3 Model 3 Summary

Scenario Highest Average Lowest Average

Fitness Fitness Fitness Time

1 88.87 87.22 85.74 21:27

(Settings 2)

2 89.35 87.66 85.94 23:32

(Settings 2)

3 89.09 87.44 85.84 24:33

(Settings 2)

The models have shown that the artificial intelligence was able to

cope up with the problem. Each settings has produced varying answer

but have shown that from the data gathered, higher settings can produce

higher quality results.

48
 CHAPTER 5

SUMMARY, CONCLUSION AND RECOMMENDATION

Summary of Findings

The usage of models was meant to test the capabilities of the

artificial intelligence. Judging by the performance of results compared to

models, the system was able to generate solutions that have at least 80%

fitness (basing on set of highest solutions per scenario). The problem lies

on the inability of artificial intelligence to perform minor corrections on

adjusting schedules (see generation results category “C”). Nevertheless,

the algorithm was able to cater for majority of the entries on the evaluation

matrix.

The lack of testing in models and the amount of model itself does

not show the exact capabilities and limitation of the system. With

generation of results from one (1) model, three (3) settings and one (1)

scenario taking almost an hour, there is only a limited amount of data that

can be used for evaluation. However, it is fair to say that the system has

performed to its capability despite of limitations imposed by the

algorithm’s configuration.

It is important to note that the generated random values for settings

was limited, therefore, the system was not able to perform at its peak. The

49
tremendous amount of possible combinations for generation of the system

have made finding the actual evaluation elusive.

Conclusion

The results have shown that the system can provide valid solutions

that can be used. However, it does not provide complete automation. There

are still scenarios that would require the operator to adjust some entries

to create a perfect solution.

The system was also designed to be simple and straightforward.

This eliminates any confusion caused by scattered user interface controls

and makes usage of the software fully utilized.

The simplicity of the system and introduction of configurable

algorithm’s goal and performance reduced the need for so much

constraints as solutions are made dynamically. This enables users to

easily use and experiment with the application until they find perfect fit

for their scenario.

The large amount of combinations for testing in order to find an

accurate evaluation for the application has proven to be far from

possibility. However, it can be concluded that from the models provided,

the system was able to generate results that despite being imperfect still

50
remains valid and acceptable given the number of constraints imposed to

it.

The solutions that the system will provide will heavily depend on

the running configuration and evaluation matrix. One may find a perfect

solution if the application was given enough time and computing power.

The complete evaluation for the system will remain hard to solve as the

freedom for the configuration of the algorithm has provided a large amount

of combinations. It can also be inferred that evaluation using other

methodologies will yield the same amount of result.

Recommendation

The researcher’s goal was to develop a solution for timetabling

problem. However, this does not mean that it should be used for every case

where more constraints are being implemented than the available ones in

the system. The researcher strongly recommends usage of other

timetabling solution for more complex and constraint needs until further

development on the existing system is available. However, usage of the

system is still promoted as it is easier to use than other solutions.

51
Bibliography

[1] “The Usage of Operations Management”, Universal Class. Retrieved

from https://www.universalclass.com/articles/business/the-usage-

of-operations-management.htm

[2] “Operations Management: Chapter 14”, IBS Center for Management

Research. Retrieved from

http://www.icmrindia.org/courseware/Operations%20Management%

202nd%20Edition/Operations%20Scheduling.htm

[3] R. Dan Reid and Nada R. Sanders, “Operations Management: Chapter

15”, 2010. Retrieved from

www.csus.edu/indiv/b/blakeh/mgmt/documents/opm101chapter15

_000.ppt

[4] Althea Gonzales and Jessy Go, “Earlier classes, longer breaks draw

mixed reactions”, June 2014. Retrieved from

http://thelasallian.com/2014/06/10/earlier-classes-longer-breaks-

draw-mixed-reactions/

[5] Anne Marxze Umil, “New school year, same old problems: K to 12,

shortages in classrooms, teachers”, June 2017. Retrieved from

http://bulatlat.com/main/2017/06/06/new-school-year-old-

problems-k-12-shortages-classrooms-teachers/

[6] Paulyn Navarrete and Illiana Tan, “Persistent issues faced during

enlistment”, January 2015. Retrieved from

http://thelasallian.com/2015/01/29/in-review-persistent-issues-

faced-during-enlistment/
[7] Careen L. Malahay and Tweeny M. Malinao, “Elementary school

shortens classes to accommodate pupils”, June 2012. Retrieved from

http://newsinfo.inquirer.net/207699/elementary-school-shortens-

classes-to-accommodate-pupils

[8] “Knapsack Problem Algorithms”, University of Colorado, Denver, May

2017. Retrieved from

http://math.ucdenver.edu/~sborgwardt/wiki/index.php/Knapsack_P

roblem_Algorithms

[9] Carsten Murawski and Peter Bossaerts, “How Humans Solve Complex

Problems: The Case of the Knapsack Problem”, October 2016. Retrieved

from https://www.nature.com/articles/srep34851/

[10] Stephen A. Cook, “An Overview of Computational Complexity”, June

1983. Retrieved from

https://dl.acm.org/citation.cfm?doid=358141.358144

[11] Paul E. Black, “NP-Complete”, December 2016. Retrieved from

https://xlinux.nist.gov/dads/HTML/npcomplete.html

[12] J.D. Ullman, “NP-Complete Scheduling Problems”, June 1975.

Retrieved from

http://www.sciencedirect.com/science/article/pii/S0022000075800

080

[13] Ahmed Wasfy and Fadi A. Aloul, “Solving the University Class

Scheduling Problem Using Advanced ILP Techniques”. Retrieved from

http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=83CF2A33

80EB4E45C5B2241036C7FAE2?doi=10.1.1.140.3740&rep=rep1&typ

e=pdf
[14] Ellis Cohen and Jarurat Ousingsawat, “An Introduction to

Algorithms for Solving Schedule-Related Problems”, January 2015.

Retrieved from http://www.project.net/introduction-algorithms-

solving-schedule-related-problems

[15] J Zhou, P E D Love, X Wang, K L Teo and Z Irani, “A review of

methods and algorithms for optimizing construction scheduling”,

March 2013. Retrieved from

https://link.springer.com/article/10.1057/jors.2012.174

[16] Melanie Mitchell, “Genetic Algorithms: An Overview”, 1995.

Retrieved from

http://ohm.ecce.admu.edu.ph/wiki/pub/Main/ResearchProjects/mit

chell_GA_tutorial.pdf

[17] David B. Fogel, “Phenotypes, Genotypes, and Operators in

Evolutionary Computing”. Retrieved from

https://pdfs.semanticscholar.org/1db9/a59ebe9b4808722fee55601b

eb8e2eda5064.pdf

[18] Khalid Jebari, “Selection Methods for Genetic Algorithms”,

December 2013. Retrieved from

https://www.researchgate.net/publication/259461147_Selection_Me

thods_for_Genetic_Algorithms

[19] A.J. Umbarkar and P.D. Sheth, “Crossover Operators in Genetic

Algorithms: A Review”, October 2015. Retrieved from

http://ictactjournals.in/paper/IJSC_V6_I1_paper_4_pp_1083_1092.p

df
[20] O. Kramer, “Genetic Algorithm Essentials, Studies in

Computational Intelligence Chapter 2”, 2017. Retrieved from

www.springer.com/cda/content/document/cda_downloaddocument/

9783319521558-c2.pdf

[21] Microsoft, “UX Checklist for desktop applications”. Retrieved from

https://msdn.microsoft.com/library/windows/desktop/dn742479.as

px
Appendix A

Letters
 Appendix B

Title Proposal
 Appendix C

College Curriculums
 Appendix D

Sample Output
 Appendix E

Program Listing
 Appendix F

Application Results
Appendix G

Gantt Chart
 Appendix H

Curriculum Vitae
Appendix I

Final Permit