The application of Models Portability to predict undergraduate students’ performance by using Transfer Learning

• Autor(a) principal: Beltran, Carlos Antonio R.
• Data de Publicação: 2022
• Outros Autores: Xavier Júnior, João Carlos, Barreto, Cephas Alves da Silveira, Gorgônio, Arthur Costa, Costa, Song Jong Márcio Simioni da
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Research, Society and Development
• Texto Completo: https://rsdjournal.org/index.php/rsd/article/view/27938
• topic: Transfer Learning; Machine Learning; Student Performance; Moodle.; Transferencia de Aprendizaje; Aprendizaje Automático; Rendimiento Estudiantil; Moodle.; Transferência de Aprendizado; Aprendizado de Máquina; Desempenho do aluno; Moodle.

One of the great challenges of education in recent years has been to accurately and reliably predict students’ performance in order to apply different strategies in order to help them with their academic deficiencies. Based on this fact, the main goal of this work is to apply a Transfer Learning approach on Learning Management Systems logs (i.e., Moodle) in order to achieve good portability of models and then predict the performance of undergraduate students. Two different scenarios have been implemented considering the activities of each course used in Moodle, the group of similar courses of the same degree as the first scenario and the group of a similar level of usage of activities as the second one. Empirical analysis has been conducted in order to evaluate the performance of the models created with three well-known classification algorithms (i.e., Decision Tree, Random Forest and Naive Bayes). AUC ROC, F-Measure, Precision and Recall have been applied as prediction measures for choosing the best models and evaluating their portability performance to the other courses. Even in the early stage, the experimental results encourage us to state that it is possible to apply transfer predictive models to the same group of courses in the majority of the cases.