The problem of student final grade prediction in a particular course has recently been addressed using data mining techniques. In this paper, we present two different approaches solving this task. Both approaches are validated on 138 courses which were offered to students of the Faculty of Informatics of Masaryk University between the years of 2010 and 2013. The first approach is based on classification and regression algorithms that search for patterns in study-related data and also data about students' social behavior. We prove that students' social behavior characteristics improve prediction for a quarter of courses. The second approach is based on collaborative filtering techniques. We predict the final grades based on previous achievements of similar students. The results show that both approaches reached similar average results and can be beneficially utilized for student final grade prediction. The first approach reaches significantly better results for courses with a small number of students. In contrary, the second approach achieves significantly better results for mathematical courses. We also identified groups of courses for which we are not able to predict the grades reliably. Finally, we are able to correctly identify half of all failures (that constitute less than a quarter of all grades) and predict the final grades only with the error of one degree in the grade scale. [For the full proceedings, see ED592609.]