Predição de evasão de cursos técnicos em EaD através de técnicas de aprendizado de máquina em duas etapas

Abstract

Educational Data Mining integrates numerous techniques that support the processing and analysis of data generated and collected from Learning Management Systems (LMS). It aims to extract relevant information in the school environment and uses Machine Learning as its main technique. The general objective of this thesis is to define a methodology that helps educational managers in detecting the risk of dropout of students in distance education, based on changes in behavior characteristics extracted by Machine Learning techniques. Due to the dynamic context of the educational environment, the proposed models are built at different times of the course, with data collected from the students’ interactions with the LMS at 10%, 25%, 50% and 75% of the two-year duration of technical courses. To provide better predictors, a two-step technique is presented, with an unsupervised approach to group students without defining a number of groups a priori, and then a supervised approach in which the cluster assigned to each student is a new input attribute to a classification model. Unsupervised techniques are also employed as a tool to study the data domain. As a result, the K-means clustering algorithm revealed the presence of four coherent groups of students according to their behavior toward the LMS, and contributes to the improvement of the prediction of students at risk of dropping out, with a return of the F1 metric above 80% in the different classifiers tested. The results show a high correlation with the completion or non-completion of the course and bring insights and new knowledge about the students. This work addresses these two techniques in cascade or in two stages, and no research with that approach has been found in Distance Education. In addition to these results, this work is relevant due to being focused on technical courses, which are considered of great importance for the social and economic development of the country, despite research and studies focusing on this level of education being scarce.