Exploring learning analytics approaches to minimize undergraduate evasion

Abstract

One of the most difficult challenges that educators face today is reducing the high student dropout rates in their institutions. Usually, the primary goal of Learning Analytics approaches in this topic is to produce a binary classification of students that are prone to drop out or not. However, this is not enough for educators to initiate a personalized intervention to reduce the evasion’s rate. Also, the structure of the curriculum plays a prominent role in the students’ performance, and despite this fact, works that analyze curricula’s structures are scarce in the literature. This dissertation proposes two approaches to minimize the evasion in the Computer Science program at the Federal University of Ceará (UFC) by analyzing data from 892 students. At first, an in-depth analysis of the acquired data to find patterns and get insights is presented. Then, we propose a prediction strategy based on the classification with reject option paradigm, in which students are classified into the two classes described above and may also reject the patterns with a high probability of being misclassified. These are probably the ones who should be subjected to an intervention. Finally, we also propose a data mining technique that evaluates a curriculum’s structure by building a linear model describing the relationship between courses based on the students performance information. The results are visualized in a user-friendly tool, which allows for contrast and comparison between the actual structure and the modeled one.