Students’ high dropout rate is a common problem on Massive Open Online Course (MOOC) platforms. Therefore, minimizing this problem requires identifying as soon as possible students who have a high probability of dropping out of the course. In this paper, we propose an approach to predict student dropout in the Virtual School of Government (EV.G), elaborating a model that can predict the students’ situation at the end of the course based on their performance in the first days and information about their access to the online platform. The processing of the data used was performed with subsequent application of automatic learning techniques, using historical data from EV.G.