University Student Desertion Analysis using Agent-Based Modeling

Approach

M. C. Castellanos Rojas, L. D. Alvarado Nieto and J. E. Villamil Puentes

Departament of Engineering, Universidad Distrital, Bogotá D.C., Colombia

Keywords: Agent-Based Model, University Student Dropout, Student Retention, National Educational Ministry (MEN).

Abstract: Student dropout at universities is a worldwide phenomenon that exceeds a 40% rate of the students admitted

to first semester. In Colombia, it exceeds 45%, an alerting rate that makes studies on the subject very

important for Governments and universities considering that it has a huge social impact and affects the

resources of the education area. Traditionally, studies are performed by statistical and mathematical

methods and the Ministry of National Education acknowledges that they have been insufficient since they

fail to explain the dropout behavior. Agent-based modeling and simulation (ABMS) has been considered a

new way of doing science when managing social problems that are complex systems forcing an evolution

simulation as a useful approach to develop this phenomenon.

1 INTRODUCTION

Students´ dropout rate at Colombian universities is

of 45.8%, a major phenomenon in every sense for

the country´s education (MEN and Permanencia

Group, 2015). The society, the Government and the

universities themselves are concerned about the

study of dropout structure and behavior. Not only

Colombia, but also countries like Mexico show

desertion rates per cohort of 40%, Argentina 43%,

Venezuela 52%, and Chile presents 54%. There are

several researches in this field, but the Colombian

Ministry of National Education -MEN- suggests that

studies are insufficient considering that adjustments

resulting from these studies do not decisively impact

the problem (MEN & Qualificar Group, 2015).

Understanding this phenomenon is difficult due

to multifaceted factors which do not allow

establishing a single structure and behavior.

In the course of his career, a student goes

through states such as: a normal performance,

dropout, low performance, subject completion and

graduation; Today, each state depends on one or

more of the following factors: academic, attitudinal,

socioeconomic, personal and institutional.

The majority of studies about these problems are

based on what has been called “traditional methods”

(statistical and mathematical), but they fail when

explaining all kind of behaviors. This paper adopts

an agent-based modeling, based partly on historical

data and meaningful narrative, making a difference

in comparison with those methods used in previous

researches.

Based on this methodology, an initial space is

composed by these agents: student, subject, teacher

and institution, and their corresponding attributes. Its

behavior is directed by very simple rules that they

modify themselves autonomously throughout time

evolution as simulation progresses.

Therefore, other ways and methods of studying

this complex phenomenon are meant to be found

through this research on agent-based simulation.

2 MAIN STUDIES

When discussing desertion, a distinction between

some studies carried out worldwide and those that

have been advanced in Colombia is made, mainly

addressed from qualitative and quantitative analyses.

Worldwide studies have focused their efforts on

defining approaches that can be classified into five

major categories such as: psychological,

sociological, economical, organizational and

interactionist as exposed by Cabrera et. al (1993). In

turn, these studies take into account explanatory

variables, which are drawn together in personal,

family or institutional.

Early researches made on the topic of dropout

were focused on sociological and psychological

128

Castellanos Rojas, M., D. Alvarado Nieto, L. and Puentes, J.

University Student Desertion Analysis using Agent-Based Modeling Approach.

DOI: 10.5220/0006777601280135

In Proceedings of the 3rd International Conference on Complexity, Future Information Systems and Risk (COMPLEXIS 2018), pages 128-135

ISBN: 978-989-758-297-4

approaches based on Durkheim´s Suicide Theory in

which he raises the university as a society, where the

student becomes part of it by interacting with others,

demonstrating their disposition, interest, attitude,

skills and expectations; Although when failure

appears, the student assumes a suicide-like behavior

(Spady, 1970).

2.1 Main Studies in Colombia

Since 2003, the Ministry of National Education

within the policy of coverage expansion, has been

developing the project " Higher Education Dropout

Decrease" with the support of the Studies on

Economic Development Center (known as CEDE by

its initials in Spanish) of the Andes University

(CEDE, 2007) through the "Research on desertion in

higher education institutions in Colombia". They

have built different strategies in order to follow up

the desertion through the System for Prevention and

Analysis of Desertion in Higher Education

Institutions (or SPADIES by its initials in Spanish),

which now has historical data of Colombian

universities in categories such as socioeconomics,

individuals, institutional and academicals variables

(MEN et al., 2009).

The District University made a specific study

about low academic performance, were the

following factors were considered: academic,

socioeconomic, aptitude and vocational, personal

and familiar, institutional, habits and study methods.

In addition, two models were raised, a linear one that

measures the academic performance and another of

logistic regression that calculates the probability of

the risk of incurring in a low yield. (Quintero,

Vásquez, Torres, Estrada, & Castellanos, 2015).

Other researches worth of mention are

"Discovery of dropout profiles with data mining

techniques" by the University of Nariño (Timarán P,

Calderón R, & Jiménez T, 2013); “Determinants of

student dropouts at the University of Antioquia”

(Vasquez V et.al. 2003); “Student Dropout at the

National Pedagogical University” (COAE, 2006);

and “A matter of survival in the National

University” (Pinto, et.al. 2007). From these studies

different definitions were adopted such as Dropout

understood as the situation in which a student incurs

when he aspires to obtain a university degree and

does not succeed; and Deserter as the person who

has been admitted to a higher education institution,

but does not register in the university for 2

consecutive semesters.

Considering the above, there are two types of

dropout, depending on the time and space according

to Vásquez et.al. (2003) and Castaño et.al. (2004).

Regarding time, as shown in Figure 1, dropout is

classified as: Premature dropout, when a person

who, having been accepted to enter the first semester

in the university, does not sign up; Early dropout,

the person who leaves its studies in the first four

semesters of career; Late dropout, refers to the

person who leaves studies in the last six semesters,

that is, from the fifth semester onwards (Vásquez V.

et al., 2003 and Castaño et al., 2004b); and

Ungraded late desertion when a person abandons its

studies once he´s finished subjects, but does not get

a graduation.

Figure 1: Classification of desertion according to time.

Adapted from: (Vásquez V et al., 2003) and (Castaño et

al., 2004).

Regarding environment, as shown in Figure 2,

dropout is divided as follows: Intern or of an

academic program: refers to the student who decides

to change his academic program by another that

offers the same university institution; Institutional as

the case in which the student leaves the university;

and educational system dropout, when the student

leaves the study. (Vásquez V et al., 2003) and

(Castaño et al., 2004)

Figure 2: Classification according to the environment.

Adapted from: (Vásquez V et al., 2003) and (Castaño et

al., 2004a).

University Student Desertion Analysis using Agent-Based Modeling Approach

129

Relevant factors for dropout were taken into

account set out by the Ministry of Nacional

Education, but only variables shown in Table 1 were

considered.

Table 1:Variables for models.

Individuals

Type of Registration

Gender

Age

Compensation

Resignation

Academics

Department

Project

State

Icfes score

Average

Kind of test

Number of academic tests

Approved subjects

Failed subjects

Academic performance

Institutional

Cohort

Entry year

Entry period

Alimentary support indicator

Type of registration

Socioeconomic

Social stratum

Birthplace

3 COMPLEX SYSTEMS

There are many definitions of complex systems, but

in general terms, systems are formed by numerous

parts or components, which interact typically in a

nonlinear way; these systems can arise and evolve

through self-organization, so that they are neither

completely regular nor completely random, allowing

the development of collective behaviors (emerging)

in macroscopic scales (Sayama, 2015).

These systems are characterized for: emergency,

self-organization, nonlinear behaviors, sensitivity to

initial conditions, they are unpredictable, do not

have a centralized control, they can present sudden

changes, managing of uncertainty, evolution,

adjustment, among others.

3.1 Agent-based Modeling

From the point of view of complexity sciences, the

study of the complex systems, given its

characteristics, requires certain particular forms and

methodological strategies to analyze the resulting

behaviors of the interaction between different

agents, which not only function in unstable

environments but also can be or not homogeneous

and in addition can be adaptive and even auto-

organize and evolve; then, a computer simulation

allows the comprehension of their structure and the

study of their behavior, but also lead to possible

emergent behaviors.

Simulation is a particular type of modeling

(representation) and introduces the possibility of a

new way of thinking about the processes social and

economic, based on ideas of the appearance of the

complex conduct of relatively simple activities

(Simon 1996), mentioned in (Gilbert and Troitzsch,

2005).

Axelrod considers a simulation of a third route of

knowledge, given that the resultant model improves

the comprehension, representation or explanation of

complex processes; " The simulation is the third way

of doing science, in contrast with the induction and

the deduction " (Axelrod, 2005).

His logic lies on the idea of a possible

reproduction of structures, behaviors or global

functions of a system, from the characterization of

its components (agents), the environment and the

local interactions agent-agent and agent-

environment (Gómez, 2016).

Some patternmakers consider any type of

independent component, already belong to software

or a model to be an agent (Bonabeau 2001). On the

other hand, other authors insist that the behavior of a

component also must be adaptive in order to be

considered as an agent. Casti (1997), mentioned in

(Macal and North, 2014).

Figure 3: Elements in Agent-based Modeling. Source:

authors.

COMPLEXIS 2018 - 3rd International Conference on Complexity, Future Information Systems and Risk

130

4 METHODOLOGY

This method is based on approaches made by

(Gilbert and Troitzsch, 2005), (Cioffi-Revilla, 2014),

(Wilensky and Rand, 2015) that agree on defining a

reference system posing a question of what you want

to resolve, considering three stages: model design,

model making and model analysis.

Model Design: precise scope, what is included

and what is possible for the model, agents and types,

attributes and behaviors, the environment, possible

events in each interval of time, input and output,

measurements and validation based on significant

abstractions. This stage can be presented through

flux diagrams, UML (unified modeling language) or

Natural Language used in this project.

Model Making: either using a specific software

or an existing package for simulation (Repast,

Netlogo, Java, Python); incremental development

with progressive iterations, from a basic to a final

model (Wilensky and Rand, 2015).

Model Analysis: verification comes first by

establishing that the simulation program works as

expected; then comes validation by analyzing the

results to asses if simulation is a good model in

accordance to the objective.

For practical reasons, this methodology was

made to be developed in two cycles, the first one

covers an agent-based simulation based on the

logistic regression model and the last one uses an

agent-based simulation based on autonomous agents.

This document addresses the first of these, using

Netlogo as tool, along with a PostgreSQL database,

to store the simulated data for each student per

semester and with output to a file in CSV format.

5 MODEL

In this case, the model base is made from 2924

records of students belonging to five careers part of

the Engineering Faculty at the District University,

with information supplied by the Advisory Office of

Systems, on cohorts from 2010-1 to 2015-3.

Included variables were: academic, personal,

socioeconomic and institutional.

Based on this information, quantity and

percentage of students were established by the

curricular project, gender, social stratum, type of

inscription, birthplace, age of entry and ICFES score

as shown in tables 2 to 7.

Table 2: Project were the student belongs.

Curricular project Frequency Percentage

Systems Engineering 579 0,20

Cadastral Engineering

and Geodesy

652 0,22

Electric Engineering 489 0,17

Electronic Engineering 577 0,20

Industrial Engineering 627 0,21

Total

2.924 1,00

Table 3: Gender of the student. Faculty of Engineering.

Gender Frequency Percentage

F 684 0,23

M 2240 0,77

Total

2.924 1,00

Table 4: Social stratum of the student.

Social stratum

Frequency Percentage

1 1017 0,35

2 1078 0,37

787 0,27

4 42 0,01

Total 2924 1,00

Table 5: Age of ingress of the student.

Age Frequency Percentage

Between 15 y 17

949 0,32

Between 18 y 20

1725 0,59

Between 21 y 25

214 0,07

More than 26

36 0,01

Total

2.924 1,00

Table 6: Type of inscription of the student.

Inscription Frequency Percentage

Normal 2.766 0,94

Minorities 119 0,04

Others

0,01

Total 2.924 1,00

Table 7: Birthplace of the student.

Birthplace Frequency Percentage

Bogotá 2.065 0,70

Surroundings 164 0,06

Outside Bogotá 434 0,15

No Data 261 0,09

Total 2.924 1,00

5.1 Model Agents

Agents were defined as: student, subject, teacher,

institution, with their respective characteristics and

rules of behavior. Nevertheless, phase one used a

base model using student and subject or matter as

University Student Desertion Analysis using Agent-Based Modeling Approach

131

agents, but further models will be worked with all

agents.

5.1.1 Student Agent

An agent student aims to travel across the curricular

map, approving the study plan to obtain the degree;

it has characteristics such as: name, curricular

project, social stratum, Icfes mark, type inscription,

birthplace, alimentary support indicator, among

others; in addition, it is subject to the rules of

advance, stop, loss, academic performance, not

repetition, that establishes the academic regulation

of the University for the student’s permanency.

Every agent student has an initial load of the total of

credits and subjects for the first semester and

changes depending on the conditions that can be: of

objective aim, of test or low performance of the

student.

Projects with the deserter's conditions: early

deserter, late deserter, late deserter without degree,

proposed in previous studies.

For the calculation of the academic performance

in students, Farmer took the following model based

on Quintero et al., (2015)

RA = 10x + 25(1- I

) + 5I

+10I

+ 10 1

1 + n

(1)

Where:

x is the student’s accumulated average

= index of repetition

= Failed subjects

Number of taken subjects

(2)

= index of permanency

= Number of admissions

Number of semesters since Entry

(3)

= index of leveling

= Number of approved subjects

Total Subjects

(4)

n = number of academic tests

6 RESULTS

Simulation allows tracking a student or groups of

students, distributed by gender and stratum

parameters or without this classification. Since 99%

of the students are stratum one, two and three, 18

simulations of systems engineering were made, six

for each stratum obtaining results shown in Table 8.

For verification and analysis purposes an

implementation with 200 students was made,

without varying initial conditions and similar results

were obtained.

As an example, actual conditions of the student

A who obtained an ICFES score of 485 points over

500 as a maximum score (97%) were considered.

On the first run, he fails in more than two subjects

and academic performance of 61.1% is obtained at

the end of the semester but improves in the

following semesters and ends the career in thirteen

admissions, with a final academic performance of

75.93 %.

Figure 4: Functionality for male gender. Source: authors.

Software Netlogo®.

Conditions: career (systems engineering), year of

ingress (2014), stratum (3), birthplace (Bogota), as

observed in the figure-4.

The output of simulation indicating the number

of graduate students versus the number of dropout

students, is also shown in Figure 5, with 21,

deserters and 29 graduate students.

Figure 5: State of male students.

In Figure-6, there are the conditions of academic

low performance of the students and the number of

times that students incurred in these is observed, that

is, all the times that the students failed for average,

failed a subject for 3 times and the times that failed

more than three subjects in the same period.

COMPLEXIS 2018 - 3rd International Conference on Complexity, Future Information Systems and Risk

132

Figure 6: Academic performance in male students.

In the figure 7, there is established the degree of

major resignation of a student when it has a low

performance to confront its academic load.

Figure 7: Compensation Vs. Resignation in male students.

Table 8 shows the final status when comparing

deserter versus graduated, in order to establish the

relationship between them.

Table 8: Simulation Results Deserters Vs. Graduates.

Sim

Dropout Graduate

Total

Number Percentage Number Percentage

1 26 0,52 24 0,48 50

2 21 0,42 29 0,58 50

3 30 0,60 20 0,40 50

4 29 0,58 21 0,42 50

5 22 0,44 28 0,56 50

6 24 0,48 26 0,52 50

7 26 0,52 24 0,48 50

8 24 0,48 26 0,52 50

9 17 0,34 33 0,66 50

10 21 0,42 29 0,58 50

11 20 0,40 30 0,60 50

12 25 0,50 25 0,50 50

13 23 0,46 27 0,54 50

14 17 0,34 33 0,66 50

15 31 0,62 19 0,38 50

16 23 0,46 27 0,54 50

17 25 0,50 25 0,50 50

18 26 0,52 24 0,48 50

Promedio 0,48 0,52

The model also allows the generation of a flat

file, which contains the record, semester to semester

of every agent-student, indicating the behavior that

this one has in each of the subjects, while the whole

mesh crosses curricular, together with the conditions

for which it passes in every period and the indicator,

if it is or not in test and for what motives. Figures 8.

7 CONCLUSIONS

According to results analysis, low-performance,

desertion and loss of student quality percentages are

much in line with the actual data provided by the

Advisory Office of Systems of the university.

Eighteen simulations using a number of students

ranging from 50 to 200, resulted with an average of

desertion of 46% and graduation rate of 46% while

students completing subjects without graduation

reach 9%.

Underperformance is causing the student to drop-

out, with the highest percentage of failure on

average, followed by the repetition of a course by

more than three times.

The highest levels of desertion are detected at the

ends of the career. In other words, whether during

the first five semesters or once subjects are

completed, but without obtaining a title, late drop

without grade, the above-mentioned Ungraded late

desertion.

Agent-based modeling and simulation allow

establishing new situations of desertion and

academic performance, as indicated in chapter six,

particularly on the causes´ mentions. Simulation

ranges fit pretty well to the actual ones.

The particular simulation of students shows that

having high initial conditions, not necessarily leads

to a high performance during his career and opposite

conditions to not lead to desertion.

The fact of simulating student by student and

students altogether, facilitates validation and

verification of parameters with very similar results,

strengthen the arguments of the desertion causes and

will surely facilitate the academic measures for

improvements.

An incoming second cycle including remaining

agents to detect new variables is on its way.

University Student Desertion Analysis using Agent-Based Modeling Approach

133

Table 9: Output data file by student.

2,65 0,44 0 120 5001 1 2,65 0,56 61,1 0 0 2 1 1 2 485 1

3,81 0,35 0,5 0 5001 2 3,23 0,65 65,1 8 5 1 1 1 2 485 1

3,5 0,32 0,67 0 5001 3 3,37 0,71 67,8 6 4 1 1 1 2 485 1

3,49 0,3 0,5 0 5001 4 3,43 0,7 68,8 7 3 1 1 1 2 485 1

3,86 0,24 0,4 0 5001 5 3,65 0,78 73,3 7 3 1 1 1 2 485 1

3,53 0,24 0,33 0 5001 6 3,59 0,79 72,8 8 0 1 1 1 2 485 1

3,72 0,25 0,29 0 5001 7 3,66 0,75 72,9 7 3 1 1 1 2 485 1

3,92 0,22 0,38 0 5001 8 3,79 0,78 75,2 5 4 1 1 1 2 485 1

3,47 0,24 0,22 20 5001 9 3,63 0,79 73,2 7 0 2 1 1 2 485 1

3,52 0,23 0,3 0 5001 10 3,58 0,83 71,7 8 4 1 1 1 2 485 1

4,02 0,22 0,18 0 5001 11 3,8 0,91 74,9 6 3 1 1 1 2 485 1

4,09 0,2 0,25 0 5001 12 3,95 0,99 77,7 6 2 1 1 1 2 485 1

3,56 0,2 0,23 0 5001 13 3,76 1 75,9 6 0 5 1 1 2 485 1

4,26 0,22 0 0 10 1 4,26 0,78 84,9 0 0 1 1 1 2 457 1

3,6 0,17 0,5 0 10 2 3,93 0,88 83,9 8 3 1 1 1 2 457 1

3,82 0,12 0,67 0 10 3 3,88 0,96 85,4 7 3 1 1 1 2 457 1

3,84 0,12 0,25 0 10 4 3,86 0,91 84,7 8 0 1 1 1 2 457 1

3,54 0,14 0,2 0 10 5 3,7 0,9 82,5 7 2 1 1 1 2 457 1

3,83 0,14 0,33 0 10 6 3,77 0,89 83,1 6 4 1 1 1 2 457 1

3,06 0,16 0,29 100 10 7 3,42 0,84 78,6 5 3 2 1 1 2 457 1

4,18 0,17 0,25 0 10 8 3,8 0,86 77,4 6 3 1 1 1 2 457 1

4,16 0,15 0,22 0 10 9 3,98 0,89 80 8 3 1 1 1 2 457 1

3,66 0,14 0,2 0 10 10 3,82 0,91 78,8 8 0 1 1 1 2 457 1

4,48 0,13 0,18 0 10 11 4,15 0,99 83,2 6 0 1 1 1 2 457 1

1,84 0,14 0,17 100 10 12 3 0,99 71,4 6 0 2 1 1 2 457 1

2,8 0,15 0,23 100 10 13 2,9 0,99 68,5 0 2 2 1 1 2 457 1

2,27 0,16 0,29 103 10 14 2,59 0,99 64,3 -1 3 2 1 1 2 457 1

4,47 0,16 0,33 103 10 15 3,53 1 73,3 -2 4 5 1 1 2 457 1

COMPLEXIS 2018 - 3rd International Conference on Complexity, Future Information Systems and Risk

134

REFERENCES

Axelrod, R. (2005). Advancing the Art of Simulation in

the Social Sciences. Gerald R. Ford School of Public

Policy, 13.

Cabrera, A. F., Nora, A., & Castaneda, M. B. (1993).

College Persistence: Structural Equations Modeling

Test of an Integrated Model of Student Retention. The

Journal of Higher Education, 64(2).

Castaño, E., Gallón, S., Gómez, K., & Vásquez, J.

(2004a). Deserción estudiantil universitaria una

aplicación de modelos de duración. Lecturas de

Economía, 60, 26.

Castaño, E., Gallón, S., Gómez, K., & Vásquez, J.

(2004b). Deserción estudiantil universitaria: una

aplicación de modelos de duración. Universidad de

Antioquia, Lecturas de economía, 39-65.

CEDE. (2007). Investigación sobre Deserción enlas

instituciones de Educación Superior en Colombia.

Cioffi-Revilla (2014). Introduction to Computational

Social Science, London, Springer

Gilbert, N., & Troitzsch, K. G. (2005). Simulation for the

social scientist. Shoppenhangers Road. Maidenhead.

Berkshire. England: McGraw-Hill Education.

Gómez, N. (2016). Simulación basada en agentes. Una

metodología para el estudio de sistemas complejos.

Centro de Innovación, Escuela de Administración, 24.

Macal, C., & North, M. (2014). Introductory tutorial:

Agent based modeling and simulation.pdf>.

Proceedings of the 2014 Winter Simulation

Conference

A. Tolk, S. Y. Diallo, I. O. Ryzhov, L. Yilmaz, S. Buckley,

and J. A. Miller, eds.

MEN, & Permanencia Group, G. (2015). Educación

Superior en Cifras Boletin, 6. Retrieved from

http://cms.colombiaaprende.edu.co/static/cache/binarie

s/articles-350451_recurso_6.pdf?binary_rand=2733

website:

MEN, & Qualificar Group. (2015). Estrategias para la

Permanencia en Educación Superior Experiencias

Significativas. Retrieved from Bogotá, Colombia:

http://www.colombiaaprende.edu.co/html/micrositios/

1752/articles-350844_pdf.pdf

Montes, I., Almonacid, P., Sebastián, G., Zuluaga, F., &

Tamayo, E. (2010). Análisis De La Deserción

Estudiantil en los Programa de Pregrado de La

Universidad EAFIT. Estudios en Economía y

Empresa. Departamento de Economía. Escuela de

Administración, 118.

Pinto, M., Durán, D., Pérez, R., Reverón, C., & Rodríguez,

A. (2007). Cuestión de supervivencia Universidad

Nacional (Vol. 1). Bogotá: Universidad Nacional de

Colombia.

Porto, A., & Di Gresia, L. (2004). Rendimiento de

estudiantes universitarios y sus determinantes. La

Revista de Economía y Estadística, Cuarta época, 42,

Quintero, R., Vásquez, D., Torres, N., Estrada, J., &

Castellanos, M. (2015). Bajo Rendimiento Académico

En La Universidad Distrital Francisco José De

Caldas (U. D. F. J. d. Caldas Ed.). Bogotá:

Universidad Distrital Francisco José de Caldas.

Sayama, H. (2015). Introduction to the Modeling and

Analysis of Complex Systems. State University of New

York at Geneseo. Geneseo, NY 14454: Open SUNY

Textbooks, Milne Library.

Spady, W. G. (1970). Dropouts from Higher Education An

Interdisciplinary Review and Synthesis.

Vásquez V, J., Castaño, E., Gallón, S., & Gómez, K.

(2003). Determinantes de la Deserción estudiantil en

la Universidad de Antioqia. Centro de Investigaciones

Económicas Universidad de Antioquia.

Wilensky, U. & Rand, W. 2015. An Introduction to Agent

Based Modeling Modeling Natural Social and

Engineered Complex Systems with NetLogo, United

States of America, Massachusetts Institute of

Technology

University Student Desertion Analysis using Agent-Based Modeling Approach

135