Title:	PROPOSAL, DEVELOPMENT AND VALIDATION OF A STRATEGY FOR PREDICTING STUDENT PERFORMANCE ON THE TÔSABENDO PLATFORM
Author(s):	Gabriel Catizani Faria Oliveira and Guilherme Tavares de Assis
ISBN:	978-989-8704-62
Editors:	Paula Miranda and Pedro Isaías
Year:	2024
Edition:	Single
Keywords:	Prediction Strategy, Prediction Models, TôSabendo Platform, Student Performance
Type:	Full
First Page:	317
Last Page:	324
Language:	English
Cover:
Full Contents:	click to dowload
Paper Abstract:	One of the potential applications of prediction models in education is in online learning. Considering online education, the gamified platform TôSabendo ("now I know" in portuguese) was created based on quizzes (question and answer games) with the aim of generating engaging experiences in Higher Education Institutions. The intention is to create a challenging environment for the player, motivating them to learn the concepts presented in each question and giving them a sense of progression in the task at hand. However, the platform currently lacks a prediction strategy using prediction models to help teachers understand, through predicted knowledge, how a particular student may perform on the platform. This understanding would be valuable for improving teaching methods and the content activities of classroom subjects, both in the traditional classroom setting and on the TôSabendo platform itself. Therefore, the goal was to propose, develop, and validate a strategy for predicting student performance on the TôSabendo platform. With the proposed and developed prediction strategy, a practical experimentation was conducted involving different prediction models and fictitious datasets. The evaluation initially assessed the model that would perform best with 10 different datasets, one for novice students and another for veterans. Subsequently, the models that achieved the best results in this first experiment go through an evaluation of different hyperparameters. Overall, after evaluating the models, decision trees yielded the most satisfactory results for both novices and veterans. By further refining this model through training with different hyperparameters, accuracy and precision results close to or equal to 100% were obtained, a value that must be analyzed and evaluated in the future due to the need to create synthetic data, which suggests a possible overfitting.
	Go Back