Artiﬁcial Intelligence Algorithms to Predict College Students’ Dropout:

A Systematic Mapping Study

Henrique Soares Rodrigues, Eduardo da Silveira Santiago,

Gabriel Monteiro de Castro Xar

a Wanderley, Laura O. Moraes, Carlos Eduardo Mello,

Reinaldo Viana Alvares and Rodrigo Pereira dos Santos

Graduate Program in Computer Science (PPGI) at the Universidade Federal do Estado do Rio de Janeiro (UNIRIO),

Keywords:

Artiﬁcial Intelligence, Machine Learning, Algorithm, Students, Dropout, College, University, Systematic

Mapping Study.

Abstract:

Higher Education Institutions (HEIs), including universities, colleges, and faculties, must develop strategies to

mitigate students’ dropout rates in undergraduate courses. This is crucial for fulﬁlling their social role, deliver-

ing high-quality professionals to society, contributing to economic development, and preventing the resource

wastage. In this context, artiﬁcial intelligence (AI) algorithms have emerged as powerful tools capable of

predicting dropout rates and identifying undergraduates at risk. This study aims to investigate and discuss the

state-of-the-art in applying AI algorithms to address students’ dropout. To achieve this objective, a systematic

mapping study (SMS) was conducted, encompassing 223 studies at ﬁrst. Finally, 23 studies were selected

for in-depth analysis to explore the effectiveness of AI algorithms in predicting students’ dropout. Further-

more, we identiﬁed key methodological design issues associated with the application of these AI algorithms,

including common features and challenges in implementing these methodologies. This study contributes by

providing practitioners and researchers with an overview of the main challenges faced by AI algorithms in

predicting students’ dropout, highlighting issues related to modeling, experimental methodology, and problem

framing.

1 INTRODUCTION

Higher Education Institutions (HEIs) aspire for their

students to undergo both academic and professional

success, as it contributes to economic growth and so-

cial justice. However, one of the most problematic

issues that HEIs face is the dropout of students (Re-

alinho et al., 2022). The deﬁnition of dropout in this

study is from Kehm et al. (Kehm et al., 2019): stu-

dents leaving their university studies before having

completed their study program and obtained a degree.

Temporary dropout due to illness or pregnancy, for

example, is not considered dropout in this context.

According to Bardagi et al. (Bardagi and Hutz,

2005), reducing the dropout rates at HEIs is not only

an educational issue, but also an economic and politi-

cal issue. The dropout reduction may have a positive

impact on students’ professional and ﬁnancial trajec-

tory, and it may reduce the waste of HEIs’ resources.

To address the student dropout issue in HEIs, artiﬁcial

intelligence (AI) algorithms have been recognized as

potential tools. They can identify students at risk of

leaving educational institutions, enabling these insti-

tutions to develop policies that support students in

continuing their studies until graduation. Therefore,

this study focuses on the use of AI algorithms to pre-

dict dropout rates and identify undergraduate students

at risk of dropping out.

The objective of this study is to identify the most

common algorithms used to predict student dropout,

the features used by these algorithms, and the typi-

cal challenges in their implementation. To do so, we

conducted a systematic mapping study (SMS) to iden-

tify and analyze the existing literature on experiments

using AI algorithms to predict dropout in HEIs, con-

tributing to an overview of this issue.

The remainder of this paper is structured as fol-

lows: Section 2 details previous literature reviews on

this topic; Section 3 presents the planning and con-

duction of this SMS; Section 4 details the results of

this SMS; Section 5 discusses the ﬁndings of this

SMS; Section 6 explores the threats to validity of the

344

Rodrigues, H., Santiago, E., Wanderley, G., Moraes, L., Mello, C., Alvares, R. and Santos, R.

Artiﬁcial Intelligence Algorithms to Predict College Students’ Dropout: A Systematic Mapping Study.

DOI: 10.5220/0012348000003636

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 16th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2024) - Volume 3, pages 344-351

ISBN: 978-989-758-680-4; ISSN: 2184-433X

SMS; and Section 7 presents ﬁnal remarks and future

work.

2 RELATED WORK

Tete et al. (Tete et al., 2022) conducted a system-

atic literature review to analyze studies related to

prediction models for student dropout from HEIs.

The authors found that the most common algorithm

is the Decision Tree. The most important features

were grouped into ﬁve categories: socioeconomic

(gender, age, professional position, income, ethnic

group), academic (grades, Grade Point Average -

GPA, frequency), psychological (learning difﬁculties,

academic life satisfaction, sociability), health (well-

being, diseases, health issues), and accessibility. This

study did not identify any academic projects or ac-

tions to decrease student dropout.

Silva and Roman (Silva and Roman, 2021) also

conducted a systematic literature review. They found

that the most analyzed features in the studies relate to

socio-demographic and academic factors, as well as

psychological and motivational variables. They also

concluded that the most frequently used algorithms

are Naive Bayes, KNN, and Random Forest (a com-

bination of several decision trees).

This study, as in the previous reviews, also in-

vestigates the most common algorithms and features

used to predict student dropout. Our contribution is

to also investigate the accuracies reached by the algo-

rithms and the most common limitations and difﬁcul-

ties faced in the implementation of such algorithms,

besides validating the previous results found in the lit-

erature.

3 RESEARCH METHOD

We performed an SMS based on Kitchenham and

Charters (Kitchenham, 2012) and Petersen et al. (Pe-

tersen et al., 2015) guidelines, which prescribe the

following phases: establish research scope, execute

search, select studies, extract data, and perform anal-

ysis. The study was documented via Parsifal

, an on-

line tool to support SMS and it is detailed in the fol-

lowing subsections.

3.1 Search Strategy and Data Source

The research question that expresses the goal of this

study was formulated following the criteria speciﬁed

https://parsif.al/

at the PIO (Population, Intervention, and Outcome),

as shown in Table 1. Therefore, the formulated re-

search question (RQ) is “How are the artiﬁcial intelli-

gence algorithms used to predict dropout rates among

higher education students?”.

Table 1: PIO structure to formulate the research question.

PIO

Population Higher Education

Institutions Dropout

Intervention Artiﬁcial Intelligence

Algorithms

Outcome Algorithms,

Difﬁculties,

Accuracies, and

Features

The desirable outcome of this research is to un-

derstand which AI algorithms are most commonly

used, which variables are used by these algorithms,

how well these algorithms can predict student dropout

in terms of accuracy, and the most common difﬁcul-

ties and limitations on the implementation of such al-

gorithms. Moreover, to expand the comprehension

of the research question, the following sub-questions

(Sub-Q) were formulated:

(Sub-Q1): What are the biggest difﬁculties in us-

ing AI to predict university dropout rates?

(Sub-Q2): How do the AI algorithms use features

to predict university dropout rates?

The sources to search by the existing studies were:

ACM Digital Library, IEEE Xplore, and Scopus.

3.2 Search String

A generic search string was created from the key-

words and their synonyms. Keywords were connected

using the AND logical operator, whereas variations

and synonyms were connected using the OR operator.

The terms of the search string were selected to con-

duct a broader search including a wide range of stud-

ies. We tested different conﬁgurations of the search

string in Scopus, which is considered the largest sci-

entiﬁc publication database that indexes the most rel-

evant publication venues. After calibrating the search

string, the ﬁnal version was:

(“higher education” OR “college” OR “graduation”

OR “university”) AND (“predict*”) AND (“artiﬁcial

intelligence” OR “AI” OR “data science” OR “deep

learning” OR “machine learning”) AND (“drop off”

OR “drop out” OR “dropout”)

Artiﬁcial Intelligence Algorithms to Predict College Students’ Dropout: A Systematic Mapping Study

345

3.3 Selection Criteria

To properly address the research question and its sub-

questions, we established selection criteria to include

studies relevant to the topic and exclude those that are

not. In this study, publication year was not deemed a

relevant criterion. The adopted selection criteria are

shown in Table 2. No criteria were set regarding the

publication date, and studies from any country were

considered acceptable.

Table 2: Selection Criteria.

Inclusion Criteria

IC1 Study describes an AI technique

for predicting dropouts in higher

education.

Exclusion Criteria

EC1 Study describes an AI technique

for predicting dropouts in ele-

mentary, high school, or massive

open online courses (MOOCs).

EC2 Duplicate study.

EC3 Study is not available for reading

and data collection (ﬁles paid for

or not made available by search

engines).

EC4 Study is not peer-reviewed.

EC5 Secondary study.

EC6 Study is not written in English.

EC7 Study is not within the topic of AI

techniques to predict higher edu-

cation dropout.

3.4 Study Selection Process

After retrieving studies from the sources, the follow-

ing ﬁlters were used to select the studies: I) title, ab-

stract, and keywords screening; II) introduction and

conclusion screening; and III) full text screening.

3.5 Data Extraction

We extracted the following data for each of the ac-

cepted studies: Study ID, reference, algorithm(s)

used, features used, algorithm accuracy, and limita-

tions of the study. The extracted data were saved in a

spreadsheet form and later used to support the discus-

sion of the SMS results.

4 RESULTS

In this section, we present the survey’s main ﬁndings.

4.1 Sources of Studies

The number of studies retrieved from each source is

described in Table 3. From the search in the chosen

sources, 223 studies were retrieved: 31 were retrieved

by IEEE Xplore, 3 by ACM Digital Library, and 189

by Scopus.

After applying the inclusion and exclusion criteria

and ﬁltering, 23 studies were selected, as shown in

Tables 4 and 5. Not all features used in the studies

are displayed in the tables. When multiple algorithms

were used in a study, the algorithm with the highest

accuracy was selected.

Table 3: Number of studies by source.

Quantity of studies by source

IEEE Xplore 31

ACM Digital

Library

Scopus 189

4.2 Filtering

The ﬁltering process is described in Figure 1.

Figure 1: Studies’ ﬁltering process.

From the set of studies retrieved, 11 were ex-

cluded because the dropout prediction focused on ba-

sic education or MOOCs. Thirteen studies were ex-

cluded due to paid access. Moreover, some studies

were about AI algorithms to predict academic perfor-

mance, not focusing on dropout risk.

4.3 Country of Origin of the Studies

The selected studies analyzed dropout behavior in

HEIs of different countries. Figure 2 describes the

number of studies conducted in each country.

We identiﬁed one study from Portugal, Hungary,

Vietnam, the United Kingdom of Great Britain and

Northern Ireland, Costa Rica, the United States of

America, Brazil, Malaysia, Saudi Arabia, Italy, and

India. We identiﬁed 2 studies each from Colombia,

Chile, and Spain. Finally, we identiﬁed 4 studies

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

346

Figure 2: Studies’ ﬁltering process.

from Peru, the country with the most studies identi-

ﬁed in this SMS. From the set of 23 selected studies,

10 were from Latin America, 7 were from Europe, 5

were from Asia, and one study was from North Amer-

ica.

4.4 Algorithms in the Studies

The percentage of each algorithm found in each study

is described in Figure 3.

Figure 3: Algorithms explored in the selected studies.

In the case of studies that compared a set of al-

gorithms, we only considered in Figure 3 the algo-

rithm with the highest accuracy; therefore, it does

not represent the total percentage for each algorithm

used in the studies. In other words, it represents only

the algorithm with the highest accuracy of each se-

lected study. After performing data extraction from

the selected studies, it was possible to answer the sub-

questions, presented in the next section.

4.5 (Sub-Q1): What Are the Biggest

Difﬁculties in Using AI to Predict

University Dropout Rates?

The most common difﬁculties and limitations are re-

lated to data availability and its small volume, they are

usually data from the authors’ HEIs afﬁliation. This

limitation causes biases in the analyses. Another lim-

itation is that data may differ in time, courses, and

different HEIs, such as the behavior of dropout rates

and student satisfaction with academic life.

By acknowledging and actively working to over-

come these challenges, higher education institutions

can harness the potential of AI algorithms to make

signiﬁcant strides in supporting student success and

retention. Collaboration among institutions and re-

searchers can facilitate the sharing of knowledge and

resources, thus creating more robust, unbiased, and

adaptable predictive models.

Two studies proposed two new algorithms to pre-

dict student dropout (S12 and S22). Both stud-

ies claimed very high accuracy for their algorithms,

which should be replicated in other datasets to con-

ﬁrm such results.

4.6 (Sub-Q2): How do AI Algorithms

use Features to Predict Higher

Institutions’ Dropout?

We found that the most commonly used variables to

predict HEI student dropout can be grouped into so-

cioeconomic (gender, age, professional position, in-

come, ethnic group), academic (grades, GPA, fre-

quency, scores at entrance exams, quantity of failed

disciplines), and psychological (satisfaction with the

academic life, sociability). The majority(11) of the

analyzed studies used academic and socioeconomic

variables, only a few used(2) psychological variables,

and none used physical health and accessibility-

related variables. Thus, we could not verify Tete et

al. (Tete et al., 2022) results.

The analyzed studies on this SMS did not explore

major differences between gender, ethnicity, and age

group on the behavior of dropout prediction. How-

ever, it does not refute the existence of differences be-

tween these social groups.

The most important factors related to college stu-

dents’ dropout are academic performance, such as

grades, GPA, attendance in class, and credits taken.

The most important external factors are the psycho-

logical state of the student, such as satisfaction with

academic life and addiction to drugs. Finally, the

most used variables in AI algorithms to predict stu-

dent dropout are related to academic performance.

5 DISCUSSION

Several researchers around the globe are investigating

AI algorithms to predict student dropout, testing algo-

rithms, such as Random Forest, Cat Boost, Logistic

Artiﬁcial Intelligence Algorithms to Predict College Students’ Dropout: A Systematic Mapping Study

347

Table 4: Results extracted from the studies.

ID Reference Algorithm Main Variables Best

Accu-

racy

Limitations

S1 (Realinho

et al., 2022)

Random For-

est

Marital status, par-

ent’s formation

N/A Bias may occur

S2 (Nagy and

Molontay,

2023)

Cat Boost Hungarian entrance

exam scores, course,

gender, age

0.84 Limited to Budapest

S3 (Osorio and

Santacoloma,

2023)

Logistic Re-

gression

Depression, drug ad-

dictions

0.80 Not mentioned

S4 (Anh et al.,

2023)

Light Gradi-

ent Boosting

Grades in subjects,

attendance in classes

0.95 Bias may occur

S5 (L

opez-

Angulo et al.,

2023)

Structural

Equation

Modeling

Satisfaction with

HEI

N/A Satisfaction with academic

life may change in time

S6 (Jimenez-

Macias et al.,

2022)

Random For-

est

Grades, Employ-

ment, credits

0.99 Few data

S7 (Gutierrez-

Pachas et al.,

2023)

CNN Grades, GPA, HDI 0.98 Unequal behaviours

S8 (Zihan et al.,

2023)

Light BPM Grades, GPA 0.93 Not mentioned

S9 (Kotsiantis

et al., 2003)

Naive Bayes Occupation, grades,

attendance on tutor-

ing

0.83 Not mentioned

S10 (Moseley and

Mead, 2008)

Decision Tree Grades, age, gender 0.94 Few data

S11 (Solis et al.,

2018)

Random For-

est

Average of Grades,

academic records

0.91 Few data

S12 (Zhang and

Rangwala,

2018)

Iterative

Logistic Re-

gression

Scores of SAT and

ACT

0.98 New proposed algorithm

S13 (Pachas et al.,

2021)

Random For-

est

Quantity of fails 0.78 Lack of data diversity

S14 (Caselli Gis-

mondi

and Ur-

relo Huiman,

2021)

Neural Net-

works

Grades, use of mo-

biles

0.87 Not mentioned

S15 (Fern

andez-

Garc

ıa et al.,

2021)

Gradient

Boosting

Not mentioned 0.72 Privacy issues

S16 (Santos et al.,

2020)

Decision Tree GPA, Entrance exam

scores

0.95 Unbalanced classes

Regression, Neural Networks, Decision Tree, Naive

Bayes, KNN, Gradient Boosting, CNN, Light Gradi-

ent Boosting, Light BPM, and SVM. Some selected

studies in this SMS tested more than one algorithm.

In such cases, this study reported the algorithm with

higher accuracy. The Random Forest algorithm is

the most frequent algorithm with better performance.

Additionally, the difﬁculties reported are mostly re-

lated to the unavailability of large data sources be-

cause most of the analyzed studies used data provided

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

348

Table 5: Results extracted from the studies.

ID Reference Algorithm Main Variables Best

Accu-

racy

Limitations

S17 (S Sani et al.,

2020)

Gradient

Boosting

Academic year, high-

school GPA, chan-

nels of admission

0.93 Not mentioned

S18 (Uliyan et al.,

2021)

Neural Net-

works

Grades, GPA 0.90 Not mentioned

S19 (Agrusti et al.,

2020)

CNN Not mentioned 0.94 Data accuracy required.

S20 (Opazo et al.,

2021)

Gradient

Boosting

Grades, GPA 0.69 Different HEIs may need

different methods

S21 (Ramirez

et al., 2022)

Random

Forests

Grades, age, gender,

academic credits

0.99 Not mentioned

S22 (Daza et al.,

2022)

Hybrid Ran-

dom Forest

and Neural

Networks

Gender, Age, Aca-

demic Credits

0.99 New proposed algorithm

S23 (Revathy

et al., 2022)

K-nearest

neighbors

Not mentioned 0.97 Not mentioned

by the authors’ afﬁliated HEIs.

To develop a more reliable AI algorithm to

predict student dropout, it is necessary to retrieve

anonymized data from several HEIs in a large data

source. However, it is a hard task to execute since

different HEIs have different data formats, such as

grades that can be expressed on a scale of 0 to 10, on a

scale from F to A, or another format and variables, by

different legislations, such as the General Data Pro-

tection Regulation (GDPR) from the European Union

(European Commission, 2016) or the General Law on

Data Protection (LGPD) from Brazil (Brasil, 2018).

Collaborative efforts among HEIs, researchers,

and regulatory bodies are essential to overcome these

challenges. Establishing data-sharing agreements that

adhere to legal requirements while facilitating the ex-

change of anonymized data for research purposes can

help unlock the potential for more reliable AI algo-

rithms. Furthermore, initiatives to create standardized

data formats and encourage transparency in data col-

lection practices can contribute to the development of

a more cohesive and effective research ecosystem fo-

cused on predicting student dropout.

The majority of the analyzed algorithms used data

related to academic performance, such as grades and

GPA, to predict student dropout, or concluded that

such categories of features are the most signiﬁcant

for making such predictions. However, it was not ex-

plored how grades are inﬂuenced by another variable.

In future work, it will be possible to investigate how

AI algorithms predict academic performance, such as

based on grades.

Another aspect to be explored is the inﬂuence

of non-academic features on academic performance.

These could include socioeconomic factors, such as

family background, ﬁnancial stability, and access to

support services. Additionally, personal factors such

as motivation, study habits, and mental health can

signiﬁcantly impact a student’s grades. Investigat-

ing how these variables interact with academic perfor-

mance can help create a more comprehensive under-

standing of the factors contributing to student dropout

risk.

Moreover, a subject of interest could be the tem-

poral aspect of academic performance prediction. An-

alyzing how students’ grades evolve and how early

warning signs in academic performance can be iden-

tiﬁed can be crucial for proactive interventions to pre-

vent dropout. Furthermore, the application of ad-

vanced AI techniques, such as machine learning in-

terpretability methods, could help shed light on how

certain features or variables contribute to academic

performance predictions. This can provide valuable

insights into the underlying mechanisms that drive the

results of AI models.

6 THREATS TO VALIDITY

The main threats to this SMS are related to the strate-

gies adopted to create the search string, retrieve pri-

Artiﬁcial Intelligence Algorithms to Predict College Students’ Dropout: A Systematic Mapping Study

349

mary studies, and extract data from these primary

studies. The completeness of this SMS may have

been affected by the missing relevant primary stud-

ies because some of them may not be retrieved by

the search string, or because some of them were ex-

cluded by EC3 due to paid access. The authors are

aware that considering only peer-reviewed studies on

the topic of using AI algorithms for predicting HEI

student dropout does not allow for the generalization

of the results, as there may be relevant content on this

topic in grey literature, such as technical reports.

In addition, the quality of this SMS may also be

inﬂuenced by potential biases introduced during the

selection and inclusion of primary studies. The cri-

teria used to determine which studies to include and

exclude could inadvertently introduce bias, affecting

the overall comprehensiveness and representativeness

of the ﬁndings.

7 CONCLUSION

We performed an SMS in which 23 studies were se-

lected for analysis. The results reveal that several

HEIs around the globe are testing algorithms to pre-

dict student dropout, trying to ﬁnd the most signiﬁ-

cant features, sharing their limitations, and trying to

maximize the algorithms’ accuracy.

From the results, we conclude that there is no spe-

ciﬁc recommended algorithm to predict higher edu-

cation students’ dropouts. Many studies test different

algorithms to perform this task, looking for the one

with the highest accuracy. In our search, the Ran-

dom Forest algorithm was the one that had a better

performance in most of the studies. The most recom-

mended features are related to academic performance,

such as grades, GPA, credits taken, and attendance in

class. Psychological health features, such as satisfac-

tion with academic life, drug addiction, and mental

diseases are also present but are less used. The most

common difﬁculties in implementing these AI algo-

rithms are related to the unavailability of a large quan-

tity of data to be used and the diversity of realities in

which different HEIs and undergraduate courses are

inserted.

Based on this study, we hope to contribute to the

ﬁeld by providing the current overview of the AI al-

gorithms used in predicting HEI students’ dropout.

ACKNOWLEDGEMENTS

This study was partly ﬁnanced by Coordenac¸

de Aperfeic¸oamento de Pessoal de N

ıvel Superior

(CAPES) and the Universidade Federal do Estado do

Rio de Janeiro (UNIRIO).

REFERENCES

Agrusti, F., Mezzini, M., and Bonavolont

a, G. (2020). Deep

learning approach for predicting university dropout: A

case study at roma tre university. Journal of e-learning

and knowledge society, 16(1):44–54.

Anh, B. N., Giang, N. H., Hai, N. Q., Minh, T. N., Son,

N. T., and Chien, B. D. (2023). An university student

dropout detector based on academic data. In 2023

IEEE Symposium on Industrial Electronics & Appli-

cations (ISIEA), pages 1–8.

Bardagi, M. and Hutz, C. S. (2005). Evas

ao universit

aria e

servic¸os de apoio ao estudante: uma breve revis

ao da

literatura brasileira. Psicologia Revista, 14(2):279–

301.

Brasil (2018). Lei nº 13.709, de 14 de agosto de 2018.

ario Oﬁcial da Rep

ublica Federativa do Brasil.

Caselli Gismondi, H. E. and Urrelo Huiman, L. V. (2021).

Multilayer neural networks for predicting academic

dropout at the national university of santa - peru.

In 2021 International Symposium on Accreditation

of Engineering and Computing Education (ICACIT),

pages 1–4.

Daza, A., Guerra, C., Cervera, N., and Burgos, E. (2022).

A stacking based hybrid technique to predict student

dropout at universities. J Theor Appl Inf Technol,

100(13):1–12.

European Commission (2016). Regulation (EU) 2016/679

of the European Parliament and of the Council of 27

April 2016 on the protection of natural persons with

regard to the processing of personal data and on the

free movement of such data, and repealing Directive

95/46/EC (General Data Protection Regulation) (Text

with EEA relevance).

Fern

andez-Garc

ıa, A. J., Preciado, J. C., Melchor, F.,

Rodriguez-Echeverria, R., Conejero, J. M., and

anchez-Figueroa, F. (2021). A real-life machine

learning experience for predicting university dropout

at different stages using academic data. IEEE Access,

9:133076–133090.

Gutierrez-Pachas, D. A., Garcia-Zanabria, G., Cuadros-

Vargas, E., Camara-Chavez, G., and Gomez-Nieto,

E. (2023). Supporting Decision-Making Process on

Higher Education Dropout by Analyzing Academic,

Socioeconomic, and Equity Factors through Machine

Learning and Survival Analysis Methods in the Latin

American Context. Education Sciences, 13(2).

Jimenez-Macias, A., Moreno-Marcos, M., Merino, P., Or-

tiz, M., and Delgado-Kloos, C. (2022). Analyzing fea-

ture importance for a predictive undergraduate student

dropout model. Computer Science and Information

Systems, 20:50–50.

Kehm, B. M., Larsen, M. R., and Sommersel, H. B. (2019).

Student dropout from universities in Europe: A re-

view of empirical literature. Hungarian Educational

ICAART 2024 - 16th International Conference on Agents and Artiﬁcial Intelligence

350

Research Journal, 9(2):147 – 164. Place: Budapest,

Hungary Publisher: Akad

emiai Kiad

Kitchenham, B. A. (2012). Systematic review in software

engineering: Where we are and where we should be

going. In Proceedings of the 2nd International Work-

shop on Evidential Assessment of Software Technolo-

gies, EAST ’12, page 1–2, New York, NY, USA. As-

sociation for Computing Machinery.

Kotsiantis, S. B., Pierrakeas, C. J., and Pintelas, P. E.

(2003). Preventing student dropout in distance learn-

ing using machine learning techniques. In Palade,

V., Howlett, R. J., and Jain, L., editors, Knowledge-

Based Intelligent Information and Engineering Sys-

tems, pages 267–274, Berlin, Heidelberg. Springer

Berlin Heidelberg.

opez-Angulo, Y., S

aez-Delgado, F., Mella-Norambuena,

J., Bernardo, A. B., and D

ıaz-Mujica, A. (2023). Pre-

dictive model of the dropout intention of chilean uni-

versity students. Frontiers in Psychology, 13:893894.

Moseley, L. and Mead, D. (2008). Predicting who will drop

out of nursing courses: A machine learning exercise.

Nurse education today, 28:469–75.

Nagy, M. and Molontay, R. (2023). Interpretable dropout

prediction: Towards xai-based personalized interven-

tion. International Journal of Artiﬁcial Intelligence in

Education.

Opazo, D., Moreno, S.,

Alvarez Miranda, E., and Pereira,

J. (2021). Analysis of ﬁrst-year university student

dropout through machine learning models: A compar-

ison between universities. Mathematics, 9:2599.

Osorio, J. and Santacoloma, G. (2023). Predictive model

to identify college students with high dropout rates.

Revista Electr

onica de Investigaci

on Educativa, 25:1–

10.

Pachas, D. A. G., Garcia-Zanabria, G., Cuadros-Vargas,

A. J., Camara-Chavez, G., Poco, J., and Gomez-Nieto,

E. (2021). A comparative study of who and when

prediction approaches for early identiﬁcation of uni-

versity students at dropout risk. In 2021 XLVII Latin

American Computing Conference (CLEI), pages 1–10.

Petersen, K., Vakkalanka, S., and Kuzniarz, L. (2015).

Guidelines for conducting systematic mapping stud-

ies in software engineering: An update. Information

and software technology, 64:1–18.

Ramirez, J. A. R., Garcia-Bedoya, O., and Galpin, I. (2022).

Maximizing student retention using supervised mod-

els informed by student counseling data. 3282:225–

239.

Realinho, V., Machado, J., Baptista, L., and Martins, M. V.

(2022). Predicting student dropout and academic suc-

cess. Data, 7(11).

Revathy, M., Kamalakkannan, S., and Kavitha, P. (2022).

Machine learning based prediction of dropout students

from the education university using smote. In 2022

4th International Conference on Smart Systems and

Inventive Technology (ICSSIT), pages 1750–1758.

S Sani, N., Fikri, A., Ali, Z., Zakree, M., and Nadiyah,

K. (2020). Drop-out prediction in higher education

among b40 students. International Journal of Ad-

vanced Computer Science and Applications, 11:550–

559.

Santos, G., Belloze, K., Tarrataca, L., Haddad, D., Bor-

dignon, A., and Brandao, D. (2020). Evolvedtree: An-

alyzing student dropout in universities. In 2020 Inter-

national Conference on Systems, Signals and Image

Processing (IWSSIP), pages 173–178.

Silva, J. and Roman, N. (2021). Predicting dropout in

higher education: a systematic review. In Anais

do XXXII Simp

osio Brasileiro de Inform

atica na

Educac¸

ao, pages 1107–1117, Porto Alegre, RS,

Brasil. SBC.

Solis, M., Moreira-Mora, T., Gonzalez, R., Fernandez, T.,

and Hernandez, M. (2018). Perspectives to predict

dropout in university students with machine learn-

ing. In 2018 IEEE International Work Conference on

Bioinspired Intelligence (IWOBI), pages 1–6.

Tete, M. F., Sousa, M. d. M., de Santana, T. S., and Silva,

S. F. (2022). Predictive models for higher education

dropout: A systematic literature review. Education

Policy Analysis Archives, 30:(149).

Uliyan, D., Aljaloud, A. S., Alkhalil, A., Al Amer, H. S.,

Mohamed, M. A. E. A., and Alogali, A. F. M. (2021).

Deep learning model to predict students retention us-

ing blstm and crf. IEEE Access, 9:135550–135558.

Zhang, L. and Rangwala, H. (2018). Early identiﬁcation

of at-risk students using iterative logistic regression.

In 2018 International Conference on Artiﬁcial Intelli-

gence in Education.

Zihan, S., Sung, S.-H., Park, D.-M., and Park, B.-K. (2023).

All-year dropout prediction modeling and analysis for

university students. Applied Sciences, 13:1143.

Artiﬁcial Intelligence Algorithms to Predict College Students’ Dropout: A Systematic Mapping Study

351