Extraction Student Dropout Patterns with Data Mining Techniques

in Undergraduate Programs

Ricardo Timarán Pereira

, Andrés Calderón Romero

and Javier Jiménez Toledo

Departamento de Sistemas,

Universidad de Nariño, Ciudad Universitaria Torobajo, Pasto, Colombia

Programa de Ingeniería de Sistemas,

Institución Universitaria CESMAG, Carrera 20ª No 14-54, Pasto, Colombia

Keywords: Extraction Patterns, Student Dropout, Data Mining.

Abstract: The first results of the research project that aims to identify patterns of student dropout from socioeconomic,

academic, disciplinary and institutional data of students from undergraduate programs at the University of

Nariño from Pasto city (Colombia), using data mining techniques are presented. Built a data repository with

the records of students who were admitted in the period from the first half of 2004 and the second semester

of 2006. Three complete cohorts were analyzed with an observation period of six years until 2011.

Socioeconomic and academic student dropout profiles were discovered using classification technique based

on decision trees. The knowledge generated will support effective decision-making of university staff

focused to develop policies and strategies related to student retention programs that are currently set.

1 INTRODUCTION

Countries in Latin America face similar challenges

regarding to how minimize the impact of student

dropout in higher education institutions. Many

institutions carry out diverse strategies such as:

educational loans, coverage increase, quality

assurance, improvement in access equity and

permanency, a better articulation with schools,

diverse offers to attend different dimensions, interest

and needs, and a better inclusion of graduates in the

work force. According to UNESCO International

Institute for Higher Education in Latin America and

the Caribbean (IESALC), countries in Latin America

have shown in 2003 an average coverage of higher

education of 28.7% and a student dropout rate above

of 50% (MEN, 2006a).

In Colombia, the education system has 277

higher education institutions, of which 81 are public

and 196 private. According to the National

Information System of Higher Education (SNIES)

coverage for 2006 was 26.1%, which is equivalent to

1,301,728 students (MEN, 2006a). One of the main

problems facing the Colombian higher education

system concerns the high dropout levels (UPN,

2005). Although recent years have been

characterized by increased coverage and new student

enrollment, the number of students which are able to

complete their higher education is not high,

suggesting that a large part of these abandoned his

studies, especially in the first semester (MEN,

2009). According to Ministry of National Education,

of every hundred students who enter at a university

about half fails to complete their academic year and

get graduation (MEN, 2009). In 2004, the dropout

was estimated at 49%. As causes of student dropout

it can be mentioned: economic and financial

constraints, poor academic performance, vocational

and professional disorientation and difficulty

adjusting to the college environment (MEN, 2006a).

The dropout carries high social and economic costs

that affect families, students, institutions and the

State (MEN, 2006b).

Data mining in education is not a new topic and

its study and implementation has been very relevant

in recent years. The use of these techniques allows,

among other things, to predict any phenomena

within the educational environment. Thus, using the

techniques offered by data mining, you can predict,

with a high percentage of confidence, the probability

of dropout of any student (Valero, 2009), (Valero et

al., 2010).

This paper describes the process of extracting

student dropout patterns from undergraduate

programs at the University of Nariño from Pasto city

(Colombia) using the classification technique based

on decision trees. Considering the stages of

knowledge discovery in databases, initially selected

136

Timarán Pereira R., Calderón Romero A. and Jiménez Toledo J..

Extraction Student Dropout Patterns with Data Mining Techniques in Undergraduate Programs.

DOI: 10.5220/0004543001360142

In Proceedings of the International Conference on Knowledge Discovery and Information Retrieval and the International Conference on Knowledge

Management and Information Sharing (KDIR-2013), pages 136-142

ISBN: 978-989-8565-75-4

 2013 SCITEPRESS (Science and Technology Publications, Lda.)

from the databases of the University of Nariño, the

socio-economic, academic, disciplinary and

institutional data of students who were admitted

between 2004 and 2006 to different undergraduate

programs, in order to make a complete follow-up to

2011, determining whether or not dropped out.

With these data, a data repository was built using

the PostgreSQL DBMS. These data were applied

stages of pre-processing and transformation in order

to obtain clean and ready data sets to apply the data

mining techniques. The first results were obtained

using the technique of classification based on

decision trees with the free data mining tool named

Weka. Finally, these results were analyzed,

evaluated and interpreted to determine the validity of

the obtained knowledge. The knowledge generated

will support effective decision-making of university

staff focused to develop policies and strategies

related to student retention programs that are

currently set.

The remain of the paper is organized as follows:

In Section 2 related works are presented. In Section

3, the methodology applied in this research is

explained in detail. Afterwards, Section 4 discusses

the obtained results and, finally, section 5 shows the

conclusions and future works.

2 RELATED WORKS

In Latin America, several research projects using

data mining to discover patterns of dropout have

been developed. At the National University of

Misiones (Argentina) conducted research on student

dropout using data mining techniques. Its main

objective was to maximize the quality of the models

to classify and group students according to their

academic characteristics and their social and

demographic factors. The study focused at students

of the Analyst in Computer Systems Program at the

Faculty of Natural Sciences. This research analyzed

data from the cohorts from 2000 to 2006 (Pautsch,

2009), (Pautsch et al., 2010).

Similarly, at the National University of the

Northeast (Argentina), a study was carried out with

the main goal of applying data warehousing and data

mining techniques, mainly based on clustering, in

order to find dropout profiles. The techniques were

applied to students of the Operating Systems Course

of the Information Systems Program. As results,

students were classified by their academic

performance and their demographic and

socioeconomic status, allowing a priori knowledge

of potential scenarios for academic success or failure

(La Red et al., 2010).

At the National University of Matanza

(Argentina) were applied data mining techniques to

evaluate academic performance and dropout of

students from the Department of Engineering and

Technological Research on data since 2003 to 2008.

The implementation of this process was performed

using MS SQL Server software to generate a data

warehouse, SPSS software for data preprocessing

and Weka (Waikato Environment for Knowledge

Analysis) to find a classifier for academic

performance to detect patterns of education dropout

(Spositto et al., 2010).

At Izúcar Technological University of

Matamoros (Mexico), proposed a research to

identify the causes for the desertion of its students.

Using the technique of classification and Weka as

data mining tool, they found relationships between

academic attributes that identify and predict the

chance of dropout. Additionally, they proposed a

tool that allows the tutor to predict the probability of

dropping out of any student at any time during your

stay at University (Valero, 2009), (Valero et al.,

2010).

At the University of Sabana, in Colombia,

conducted a research project where the goal was to

select, from a database of students, attributes that

have the greatest impact on the dropout of the

university in the past four years. The study used

classification by rough sets as data mining technique

using the Rose2 software package (Restrepo and

López, 2008).

Pinzon (2011) presents the characterization of

student profile defector from the School of

Marketing and Advertising at the Sergio Arboleda

University (Colombia). The author applied

clustering, using the K-means algorithm, as data

mining technique. During the study, it was analyzed

demographic variables obtained in the last student

enrollment record and the causes that generated the

withdrawal. The end result is three types of clusters

or groups of students. For the case of the research,

they were significant profiles.

3 STUDENT DROPOUT

PATTERNS DISCOVERY

PROCESS

Knowledge Discovery in Databases (KDD) is the

extraction of implicit, previously unknown, and

potentially useful information from data (Witten and

Frank, 2000). KDD is basically an automatic process

ExtractionStudentDropoutPatternswithDataMiningTechniquesinUndergraduatePrograms

137

that combines discovery and analysis. The KDD

process is interactive and iterative, involving

numerous steps with many decisions made by user.

This process usually involves preprocessing the data,

make data mining and visualize the results (Agrawal

and Srikant, 1994), (Chen et al., 1996), (Piatetsky-

Shapiro, Brachman and Khabaza, 1996), (Han and

Kamber, 2001).

In the process of discovering student dropout

patterns in undergraduate programs, the following

steps were performed:

3.1 Selection Step

The main goal of this step is selecting a data set

from internal or external sources of data, or focusing

on a subset of variables or data samples, on which

discovery is to be performed. Internal sources were

selected from the Grades and Register-UDENAR

databases from the Admissions and Academic

Control Office (OCARA) of the University of

Nariño. Given the observation window for this study

(2004-2011), these databases store personal and

academic information of 15.875 students (OCARA,

2011). As main external sources of data were

selected diverse databases from different Colombian

Institutions such as: the Colombian Institute for the

Development of Higher Education (ICFES), the

National Bureau of Statistics (DANE), the Dropout

Prevention in Higher Education System (SPADIES),

the Potential Beneficiaries of Social Programs

Information System (SISBEN) and the Colombian

National Registry of Civil Status (all acronyms come

from their names in Spanish).

From 15.875 records previously selected only

data of students of cohorts 2004, 2005 and 2006

with the attributes most relevant to this study were

chosen. The outcome is 6870 records and 62

attributes belonging to socio-economic, academic,

and institutional data. These data were stored in a

table named T6870A62 in a database called

UDENAR_REPOSITORY using PostgreSQL. This

table will be the basis for subsequent phases of the

dropout patterns discovery process.

3.2 Preprocessing Step

The goal at this stage is to obtain clean data, i.e.

data without null or outlier values, in order to

retrieve high quality patterns. Through ad-hoc

queries or histograms on the T6870A62 table, the

quality of the data available for each of its attributes

was thoroughly analyzed.

Considering the relevance of certain attributes

for this research, null values of these attributes were

updated with the values found in external sources.

However, the attributes with a high percentage of

nulls data (more than 80%), were eliminated by the

inability to obtain these values from external

sources, using statistical techniques such as mean,

median and mode or deriving their values through

others.

As result of this stage and in order to generate

knowledge about the socioeconomic, academic,

disciplinary and institutional factors, the 31 most

representative attributes were selected from

T6870A62 table. A new table was created and

called as T6870A31. From these 31 attributes, 18

were chosen to analyze the socioeconomic factor

and 15 for the academic factor. Similarly, two new

tables (T6870A18 and T6870A15) were created

respectively. A detail description of these new

tables is shown in Table 1. Given the small number

of selected attributes for disciplinary and

institutional factors, these were added in the

academic one.

Table 1: New tables of UDENAR-REPOSITORY

database.

TABLE DESCRIPTION

T6870A31

Table containing 6870 students

admitted in 2004-2006 and with 31

attributes to be considered in the study.

T6870A18

Table of 6870 students and 18 attributes

to consider social and economic factors.

T6870A15

Table of 6870 students and 15 attributes

to consider for academic factors.

3.3 Transformation Step

Data transformation includes any process that

modifies the form of the data. The aim of this stage

is to transform the data source in a dataset ready to

apply any of the different techniques of data mining.

Among the operations performed to transform the

data are: elimination of the least relevant attributes,

creation of new attributes by deriving them from

others (keeping or replacing these attributes) and / or

modification of the type of attributes (using

discretization or continuity methods).

In order to facilitate patterns extraction, the

numerical values of attributes in table T6870A31

were translated to nominal values. This process

(known as discretization) was carried out using the

discretize filter in Weka with the equal frequency

parameter (useEqualFrequency) set to 6 values.

Moreover, the T6870A31 table was adapted to

ARFF format (Attribute Relation File Format)

KDIR2013-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval

138

required by Weka to continue with the data mining

phase. Table 2 shows the attributes of the table

T6870A31 with the new discretized values.

According to this table, the first 17th attributes

together with the attribute 31th, form the T6870A18

table and correspond to the socioeconomic

attributes. Similarly, the attributes from 17th to 31st

correspond to the academic attributes and conform

the T6870A15 table.

Table 2: Attributes T6870A31.

ATTRIBUTES AND VALUES

1 gender {male,female}

2 marital_status {single,married,divorcée,civil

union,single mother,widower,religious}

3 place_of_birth

{pasto,north,south,west,center,cost,putumayo,ot

her}

4 place_of_provenance

{pasto,north,south,west,center,cost,putumayo,ot

her}

5 health_regimen {contributory, subsidized}

6 economic_status {0,1,2,3,4,5,6,99}

7 father_alive {true,false}

8 father_occupation { various, operational, art,

industry, building, professional, ...}

9 mother_alive {true,false}

10 mother_occupation { various, operational, art,

industry, building, professional, ...}

11 residence_type {leased, owner,mortgage, ...}

12 live_with_family{true, false}

13 siblings_at_university {true, false}

14 family_income{'from 4540000 to

5980000','greater than 8540000','from 2850000

to 4540000','from 5980000 to 8854000','less

than 2850000'} *figures in COP.

15 school_tuition_fee {'from 76639 to

106100','from 60248 to 76639','greater than

106100','less than 21550','from 21550 to

44369','from 44369 to 60247'} *figures in COP.

16 college_ tuition_fee {'less than 100259','from

120574 to 158846','from 100259 to

120574','from 234266 to 381504','from 158846

to 234266','greater than 381504'} *figures in

COP.

17 age_at_enrollment {'18','less than 18','greater

than 22','from 21 to 22','19','20'}

18 school_type {public,private}

19 school_day-part

{morning,afternoon,full,night,weekend}

20 icfes_weighted{'from 52 to 54','from 50 to

52','from 54 to 58','from 46 to 50','greater than

58','less than 46'}

Table 2: Attributes T6870A31. (Cont.)

21 icfes_average{'from 53 to 56','from 48 to

50','from 46 to 48','from 50 to 53','less than

46','greater than 56'}

22 icfes_total {'greater than 475','from 420 to

450','from 450 to 475','from 400 to 420','from

375 to 400','less than 375'}

23 campus

{Pasto,Tumaco,Tuquerres,Ipiales,Samaniego,Bu

esaco,'La Union',Ricaurte}

24 faculty {'natural

sciences','humanities','agricultural

sciences','economic sciences','health

sciences',engineering, ...}

25 area_program {mathematics,'social

sciences',agronomy, economy, accounting, arts,

education}

26 grade_average {'from 2.4 to 3.1','from 3.5 to

3.7','greater than 4.0','from 3.7 to 4.0','from 3.1

to 3.5','less than 2.4'}

27 failed_courses {'from 3 to 4','greater than

9','from 5 to 6',none,'from 1 to 2','from 7 to 9'}

28 failed_semester {initial semesters, middle

semesters, final semesters, not applicable,

elective course}

29 area_course {pedagogy,philosophy,history,basic

sciences, statistics, ...}

30 failed_times {'2','3',none,'1','4','greater than 4'}

31 dropout {true, false}

3.4 Data Mining Step

The goal of the data mining step is the search and

discovery for unexpected and interesting patterns

from data. In this process, intelligent task are applied

such as classification (Quinlan, 1986), (Wang et al.,

1998), (Witten and Frank, 2000), clustering (Ng and

Han, 1994), (Zhang et al., 1996), sequential patterns

(Agrawal and Srikant, 1995), associations (Agrawal

and Srikant, 1994) among others. The data mining

task chosen for the process of discovering student

dropout patterns was classification using a decision

trees technique.

Data classification provides results from a

supervised learning process. Data classification is a

two-step process. In the first step, a model is built

describing a predetermined set of data classes. The

input data, also called the training set, consists of

multiple examples (records), each having multiple

attributes or features and tagged with a special class

label. In the second step, the model is used to

classify future test data for which the class labels are

unknow (Han and Kamber, 2001).

Decision tree classification is the most popular

ExtractionStudentDropoutPatternswithDataMiningTechniquesinUndergraduatePrograms

139

model, because it is simple and easy to understand

(Han and Kamber, 2001), (Sattler and Dunemann,

2001), (Timarán and Millan, 2006). A decision tree

is a flow-chart-like tree structure, where each

internal node denotes a test on an attribute, each

branch represents an outcome of the test, and leaf

nodes represent classes. The top-most node in a tree

is the root node (Han and Kamber, 2001).

A decision tree classifier is built in two phases: a

growth phase and a pruning phase. In the growth

phase, the tree is built by recursively partitioning the

data until all members belong to the same class. In

the pruning phase, many branches are removed with

the goal of improving classification accuracy on data

(Wang et al., 1998).

The classification rules were obtained with the

Weka data mining tool (Waikato Environment for

Knowledge Analysis), using the J48 algorithm,

which implements the algorithm C.45 (Quinlan,

1993), and utilizing data repositories described in

Table 1.

The J48 algorithm is based on the use of the gain

ratio criterion. Consequently, those variables with

greater number of different values do not have

benefit in the selection. Furthermore, the algorithm

incorporates a classification tree pruning once it has

been induced (Hernández and Lorente, 2009).

T6870A31 repository was used to discover

general patterns that affect the student dropout. The

dropout attribute was chosen as the class. Similarly,

the datasets T6870A15 and T6870A18 were used to

determine, respectively, socioeconomic and

academic factors that influence student dropout.

3.5 Evaluation Step

The objective of this final stage is the evaluation and

interpretation of the obtained results in order to

consolidate the discovered knowledge with two

goals in mind. First, to integrate it into other

systems for further action, and second, to compare it

with previously discovered knowledge.

The 10-fold cross validation method was used in

order to evaluate the quality and prediction accuracy

of the discovered patterns. The most relevant

classification rules are shown in the following

section.

4 RESULTS AND DISCUSSION

As a result of interpreting the decision tree generated

by the algorithm J48 with data from T6870A31, the

most representative classification rules are shown in

Table 3. All of them have a confidence threshold

greater than 80%.

According to the rules of Table 3, the

predominant factors in the student dropout from the

University of Nariño are academics, specially a low

average in grades and the number of courses lost in

the initial semesters of the program.

In order to determine the socioeconomic factors

affecting student dropout, a number of classification

rules, with confidence greater than 80%, were

generated using the T6870A18 dataset. The results

show that the most significant socioeconomic factors

affecting student dropout are a tuition fee greater

than COP$ 381,504 (around USD$ 212) and a

provenance from the south of the department of

Nariño (Colombia). The fact of being single, living

with mother and be in the city of Pasto may also

impact education dropout.

Table 3: Most representative classification rules from T6870A31 dataset.

Classification rules Dropout class Support Confidence

grade_average = from 2.4 to 3.1 & failed_semester = Initial

semesters True 0.1559 0.939

grade_average = from 3.7 to 4.0 & failed_times = 1 False 0.1551 0.8528

grade_average = less than 2.4 True 0.1519 0.998

grade_average = from 3.5 to 3.7 & campus = PASTO &

failed_courses = from 7 to 9 False 0.0314 0.8585

grade_average = from 3.1 to 3.5 & failed_courses = from 3 to

4 True 0.0264 0.9535

grade_average = from 3.5 to 3.7 & failed_courses = De 1 to 2

& failed_semester = Initial semesters True 0.0227 0.8108

grade_average = from 3.1 to 3.5 & failed_courses = De 5 to 6

& failed_semester = Initial semesters True 0.017 0.8198

grade_average = from 3.5 to 3.7 & campus = PASTO &

failed_courses = from 1 to 2 & failed_semester = Initial

semesters & place_of_provenance = PASTO True 0.0129 0.8341

KDIR2013-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval

140

To determine other factors associated with

academic dropout, classification rules were

generated with T6870A15 dataset with a confidence

greater than 80%, without the attribute

grade_average. The results shown the factors that

influence academic dropout, in addition to a low

average of grades and the courses lost in initial

semesters, are: the faculty to which the student

belongs, specifically the faculties of Natural

Sciences, Health Sciences, Education and Arts;

Also, the area of the course which was lost, such as

the area of mathematical foundations, introduction to

natural science, basic training, pedagogy, economics

and accounting; and the campus of the University,

particularly those located in Ipiales and Tumaco

cities.

5 CONCLUSIONS AND FUTURE

WORKS

Initial results obtained through the decision tree

classification technique indicates that it is able to

generate models consistent with observed reality and

theoretical support, based only on the data that is

stored in the database, for the study case of the

University of Nariño. One of the great difficulties

faced in these kinds of studies is the poor data

quality. Often, when the cleaning process was

ended, many variables become useless by the

inability to obtain their correct values.

Unfortunately, it has a direct influence on the results

of data mining.

A set of general patterns for student dropout has

been obtained. It is mainly determined by factors

such as a low average in grades and the number of

courses students have failed at initial semesters. In

addition, socioeconomic and academic factors

related with student dropout have been identified as

well. The assessment, analysis and utility of these

patterns will support effective decision-making of

university staff focused to develop policies and

strategies related to student retention programs that

are currently set.

As future works it can be mentioned additional

studies of student dropout at the University of

Nariño using other data mining techniques such as

clustering and association rules in order to determine

affinities, similarities and relationships between

socioeconomic and academics factors of students

who drop out. To verify the quality and accuracy of

the rules obtained will be used other classifiers.

Applying the same methodology for student

dropout at CESMAG University Institution and

analyze and evaluate the patterns found in both

higher education institutions.

ACKNOWLEDGEMENTS

This research is funded by the Ministry of Education

in Colombia and counterpart funds from the

University of Nariño and CESMAG University

Institution.

REFERENCES

Adamo, J. M., 2001. Data Mining for Association Rules

and Sequential Patterns: Sequential and Parallel

Algorithms. New York (USA): Springer-Verlag. 253 p.

ISBN: 0-387-95048-6.

Agrawal, R., Srikant, R., 1994. Fast Algorithms for

Mining Association Rules. In: 20th International

Conference on Very Large Data Bases, VLDB 1994,

(12-15/09/1994). Santiago de Chile (Chile):

VLDB.Proceedings. p. 487-499. ISBN: 1-55860-153-8.

Agrawal, R., Srikant, R., 1995. Mining Sequential

Patterns. In: The 11th International Conference on

Data Engineering ICDE, pp 3-14. Taipei, Taiwan.

Chen, M., Han, J., Yu, P., 1996. Data mining: An

overview from database perspective. In: IEEE

Transactions on Knowledge and Data Engineering.

Vol. 8, No. 6 (dic). Los Alamitos (CA, USA): IEEE

Computer Society. p. 866-883. ISSN: 1041-4347.

Fayyad, Usama, Piatetsky-Shapiro, Gregory, Smyth,

Padrahic, 1996. The KDD process for extracting useful

knowledge from volumes of data. In: Comunications of

the ACM. Vol 39, No. 11 (nov). New York (USA):

ACM Digital Library. p 27-34. ISSN: 0001-0782.

Garcia Morate, Diego (s.f.). Manual de Weka (on líne),

http://www.metaemotion.com/diego.garcia.morate/do

wnload/weka.pdf (consulta: 15/06/ 2012).

Han, Jiawei, Kamber, Micheline, 2001. Data Mining:

Concepts and Techniques. San Francisco (CA, USA):

Morgan Kaufmann Publishers, Academic Press. 550 p.

ISBN: 1-55860-489-8.

Hernández, O. J., Ramírez, Q. M., Ferri, R. C., 2005.

Introducción a la Minería de Datos. Madrid (España):

Pearson Prentice Hall. 656 p. ISBN: 84-205-4091-9.

Hernández, E., Lorente, R., 2009. Minería de datos

aplicada a la detección de Cáncer de Mama.

Universidad Carlos III, Madrid.

http://www.it.uc3m.es/jvillena/irc/practicas/08-9/14.pdf.

La Red, David, Acosta, J. C., Cutro, Luis, Uribe, V. E.,

Rambo, A. R., 2010. Data Warehouse y Data Mining

Aplicados al Estudio del Rendimiento Académico. In:

Novena Conferencia Iberoamericana en Sistemas,

Cibernética e Informática, CISCI 2010, (29/06-

2/07/2010), Orlando (Florida, USA): International

Institute of Informatics and Systemics. Memorias CISCI

2010, Volumen I, p. 289-294. ISBN: 978-1-934272-94-7.

ExtractionStudentDropoutPatternswithDataMiningTechniquesinUndergraduatePrograms

141

MEN, 2006a. América Latina piensa la deserción. En:

Boletín informativo Educación Superior. No 7 (dic).

Bogotá (Colombia): Ministerio de Educación

Nacional. p 14. ISSN: 1794-2446.

MEN, 2006b. Deserción estudiantil: prioridad en la

agenda. En: Boletín informativo Educación Superior.

No 7 (dic). Bogotá (Colombia): Ministerio de

Educación Nacional. p 1. ISSN: 1794-2446.

MEN, 2009. Deserción estudiantil en la educación

superior colombiana: metodología de seguimiento,

diagnóstico y elementos para su prevención. Bogotá

(Colombia): Ministerio de Educación Nacional. 158 p.

ISBN: 978-958-691-366-9.

OCARA, 2011. Datos de estudiantes matriculados en los

programas de pregrado de la Universidad de Nariño en

el periodo A de 2004 hasta el periodo A de 2011.

Oficina de Control y Registro Académico de la

Universidad de Nariño. Pasto (Colombia).

Pautsch, Jesús, 2009. Minería de datos aplicada al análisis

de la deserción en la Carrera de Analista en Sistemas

de Computación. Tesis de grado (Licenciado en

Sistemas de Información). Posadas, Misiones

(Argentina): Universidad Nacional de Misiones. 193 p.

Pautsch, Jesús, La Red, David, Cutro, Luis, 2010. Minería

de datos aplicada al análisis de la deserción en la

Carrera de Analista en Sistemas de Computación (on

líne). Posadas, Misiones (Argentina): Universidad

Nacional de Misiones.

http://www.dataprix.com/files/Analisis%20de%20Des

ercion%20Univ_0.pdf. (consulta: 18/06/2012).

Pinzón, Liza, 2011. Aplicando minería de datos al

marketing educativo. In: Revista Notas de Marketing.

No 1 (jun). Bogotá (Colombia): Universidad Sergio

Arboleda, Escuela de Marketing y Publicidad. p 45-61.

Quinlan, J. R., 1986. Induction of Decision Trees. In:

Machine Learning Journal, Vol. 1, No. 1, Kluwer

Academic Publishers, pp. 81-106. Boston, USA.

Quinlan, J. R., 1993. C4.5: Programs for Machine

Learning. San Francisco (CA, USA): Morgan

Kaufmann Publishers. 299 p. ISBN: 1-55860-238-0.

Restrepo, Mauricio, López, Andrés, 2008. Uso de la

metodología Rough Sets en un modelo de deserción

académica. In: XIV Congreso Ibero Latinoamericano

de Investigación de Operaciones, CLAIO 2008, (9-

12/09/2008), Cartagena (Colombia). Universidad del

Norte. Libro de Memorias CLAIO 2008, p.108-109.

Ediciones Uninorte.

Sattler, K., Dunemann, O., 2001. SQL Database Primitives

for Decision Tree Classifiers. In: The 10th ACM

International Conference on Information and

Knowledge Management - CIKM, (5-10/11/2001),

Atlanta, Georgia (USA): ACM. Proceedings, p. 379-

386. ISBN: 1-58113-436-3.

Spositto, Osvaldo, Etcheverry, Martín, Ryckeboer, Hugo,

Bossero, Julio, 2010. Aplicación de técnicas de

minería de datos para la evaluación del rendimiento

académico y la deserción estudiantil. In: Novena

Conferencia Iberoamericana en Sistemas, Cibernética

e Informática, CISCI 2010, (29/06-2/07/2010),

Orlando (Florida, EE.UU.). International Institute of

Informatics and Systemics. Memorias CISCI 2010,

Volumen I. ISBN: 978-1-934272-94-7.

Timarán, R., Millán, M., 2006. New Algebraic Operators

and SQL Primitives for Mining Classification Rules.

In: proceedings of The Five IASTED International

Conference on Computational Intelligence (CI 2006),

International Association of Science and Technology

for Development, ISBN No 0-88986-603-1. San

Francisco, USA.

Timarán Pereira, Ricardo, 2009. Una mirada al

descubrimiento de conocimiento en bases de datos. In:

Ventana Informática. No 20 (ene-jun., 2009).

Manizales (Colombia): Centro de Investigaciones,

Desarrollo e Innovación, Facultad de Ingeniería,

Universidad de Manizales. p 39-58. ISSN: 0123-

9678.

UPN, 2005. La deserción estudiantil: reto investigativo y

estratégico asumido de forma integral por la UPN (on

line). In: Encuentro Internacional sobre Deserción en

Educación Superior: experiencias significativas (17-

18/05/2005) Bogotá (Colombia): Ministerio de

Educación Nacional. http://www.mineducacion.

gov.co/1621/ articles-85600_Archivo_pdf3.pdf.

(consulta: 15/06/ 2012).

Valero, Sergio, 2009. Aplicación de técnicas de minería de

datos para predecir la deserción (on línea). Izúcar de

Matamoros, Puebla (Mexico): Universidad

Tecnológica de Izúcar de Matamoros.

http://www.utim.edu.mx/~svalero/docs/MineriaDeserc

ion.pdf. (consulta: 10/06/2012).

Valero, Sergio, Salvador, Alejandro, García, Marcela,

2010. Minería de datos: predicción de la deserción

escolar mediante el algoritmo de árboles de decisión y

el algoritmo de los k vecinos más cercanos (on line)

Izúcar de Matamoros, Puebla (Mexico): Universidad

Tecnológica de Izúcar de Matamoros. http://

www.utim.edu.mx/~svalero/docs/e1.pdf. (consulta:

10/06/2012).

Wang, M.; Iyer, B., Scott, V., J., 1998. Scalable Mining

for Classification Rules in Relational Databases. In:

International Database Engineering and Application

Symposium, IDEAS 98, (08-10/07/1998), Cardiff

(Wales,U.K.): IEEE Computer Society. Proceedings,

p. 58-67. ISBN: 0-8186-8307-4.

Witten, I. and Frank, E., 2000. Data Mining: Practical

Machine Learning Tools and Techniques with Java

Implementations. Morgan Kaufmann Publishers, 365

p, ISBN: 1-55860-552-5. San Francisco, CA, USA.

Zhang, T., Ramakrishnan, R., Livny, M., 1996. BIRCH:

An Efficient Data Clustering Method for Very Large

Databases. In ACM SIGMOD International

Conference on Management of Data, pp. 103-114.

Montreal, Canada.

KDIR2013-InternationalConferenceonKnowledgeDiscoveryandInformationRetrieval

142