Skip to the main content

Original scientific paper

https://doi.org/10.15516/cje.v27i1.5950

Leveraging Metaheuristic Optimized Classifier Exploitability to Detect and Understand Student Dropout

Žaklina Spalević ; Singidunum University Belgrade, Faculty of Tourism and Hospitality Management
Lazar Stošić orcid id orcid.org/0000-0003-0039-7370 ; University UNION Nikola Tesla Belgrade, Faculty of Informatics and Computer Science
Nebojša Bačanin Džakula orcid id orcid.org/0000-0002-2062-924X ; Singidunum University Belgrade, Faculty of Informatics and Computing
Luka Jovanović ; Singidunum University Belgrade, Faculty of Informatics and Computing
Filip Marković ; University of Pristina in Kosovska Mitrovica, Faculty of Technical Sciences


Full text: english PDF 12 Kb

page 85-127

downloads: 75

cite

Full text: croatian pdf 1.422 Kb

page 85-127

downloads: 208

cite


Abstract

Education is crucial for social progress, fostering knowledge, skills, and cognitive development, and is closely linked to long-term economic growth. By enhancing the general knowledge and skills of the population, education fuels innovation and the adoption of new ideas, making human capital a key driver of economic development. A well-educated workforce significantly contributes to research and development, integrating innovations into production and promoting economic growth. Despite extensive research, maintaining high enrolment and low dropout rates remains challenging. The potential of artificial intelligence (AI) in addressing this issue is underexplored. This work aims to fill this gap by leveraging metaheuristic optimized deep neural networks to detect students at risk of dropping out. A modified version of the firefly algorithm (FA) is introduced specifically to meet the demands of this optimization. Additionally, explainable AI (XAI) techniques are employed to better understand the factors influencing student decisions, thereby aiding in the formulation of effective retention policies. The introduced methodology is evaluated on a realworld dataset, with the best models achieving an accuracy exceeding 82% for dropout detection.

Keywords

Academia; Deep Neural Networks; Explainable Artificial Intelligence; Firefly Algorithm; Legal regulation and Ethical Principles

Hrčak ID:

333387

URI

https://hrcak.srce.hr/333387

Publication date:

28.3.2025.

Article data in other languages: croatian

Visits: 686 *