Forecasting academic performance is a challenging task that involves multiple factors and dimensions, such as cognitive, affective, and behavioral aspects of students. In this paper, we analyze several approaches used to predict student outcomes based on multiple sources of data like historical performance and behavior among others. We also propose a new method to analyze facial expressions that will be used later on a model to predict academic performance. This new method uses artificial intelligence techniques, such as deep neural networks, to classify 7 different emotions and we evaluated it using a public dataset that helped us to measure its performance and improve it using data management techniques like augmentation. Our results show that our method outperforms the baseline used and provides a better understanding of the emotions depicted in each one of the images used, this will help us to continue our study on academic performance prognosis.