Análise da curva de aprendizagem de alunos e capacidade preditiva de modelos no ensino da programação

Abstract

Learning algorithms requires knowledge of fundamentals such as abstraction, decomposition and pattern identification. The learning process is modular with steps that are well defined and prerequisites for each other, which makes it difficult for the teacher to identify the student’s needs. In addition, a manual correction of students’ tasks is prohibitive for the teacher because of time restrictions. These problems are maximized for the class with many students. An alternative to alleviate such problems is the use of Intelligent Tutoring Systems. The modeling and prediction of knowledge is a fundamental part of an intelligent tutor, since it is through this information that the tutor makes his decisions and adapts the students’ learning process. This work proposes to present a qualitative analysis between the knowledge models of students in the domain of programming aiming to identify the best approaches for modeling and the peculiarities of the domain. One of the contributions of this work was the collection of data in the programming domain for the creation of the COSMODS database. In addition, an analysis of the learning curve and predictive capacity of the models generated through the proposed methodology was made. The analysis showed that the compilation errors directly influence the representation of students’ knowledge and predictive capacity of the models. Finally, there was a discussion of the impacts that compilation errors have on the generated models.