Uma abordagem com learning analytics e séries temporais na análise de dados educacionais

Abstract

The mode of distance education, previously discriminated by students from a wide range of social segments, has been established as an excellent alternative to traditional education. Its evolution has close ties with advances in information technology and communications. The expansion of broadband to the most distant places in the country, offering fast access to the Internet, favors the dissemination of distance education courses offered by private and public companies. This paradigm shift in education has brought about transformations in the behavior of managers and teachers, who become increasingly dependent on the use of technology to create more interesting and interactive didactic content, as well as on student behavior, which should adapt to the newness, from being passive agents in the educational process to becoming active agents of their own learning through self-regulating behaviors. In order for online interaction between students and teachers to occur, it is necessary to implement a virtual learning environment (VLE), such as Moodle. This environment is fundamental for communication between the actors of the e-learning, storing in their database all the interactions that students, teachers and tutors perform during the activities online. Such interactions have become a fertile field for educational data mining researchers and learning analytics to study the behavior of these students through the attributes derived from these interactions. In this context, this research presents an approach of unsupervised learning of machines, through the algorithm of k-means clustering, to discover patterns of engagement behaviors and procrastination of students of a e-learning graduation course. Student and teacher interactions were extracted from Moodle log files, VLE used by the Institution of Higher Education that offers the course, being transformed into attributes used in the creation of the time series that compose the data set of input data of the clustering algorithm. Finding as results groups of students with low, intermediate and high levels of engagement that present a relationship between procrastination behavior and performance at the end of the course.