Predicting the Risk of Academic Dropout With Temporal Multi-Objective Optimization

In the European academic systems, the public funding to single universities depends on many factors, which are periodically evaluated. One of such factors is the rate of success, that is, the rate of students that do complete their course of study. At many levels, therefore, there is an increasing interest in being able to predict the risk that a student will abandon the studies, so that (specific, personal) corrective actions may be designed. In this paper, we propose an innovative temporal optimization model that is able to identify the earliest moment in a student's career in which a reliable prediction can be made concerning his/her risk of dropping out from the course of studies. Unlike most available models, our solution can be based on the academic behavior alone, and our evidence suggests that by ignoring classically used attributes such as the gender or the results of pre-academic studies one obtains more accurate, and less biased, models. We tested our system on real data from the three-year degree in computer science offered by the University of Ferrara (Italy).