A positive emotional state of students has proved to be essential for favouring student learning, so this paper explores the possibility of obtaining student feedback about the emotions they feel in class in order to discover emotion patterns that anticipate learning failures. From previous studies about emotions relating to learning processes, we compiled a Twelve Emotions in Academia Model, which is composed of six positive and six negative emotions. Using this model as a base we built the "EmotionsModule" in the "PresenceClick" system, allowing students to identify their emotions and follow their evolution by means of visualizations. Instructors can also view an anonymized version of these data to increase their knowledge about the emotional state of the group and propose new learning strategies to improve the group's overall state. Information about attendance in face-to-face sessions has been also considered due to the fact that it is positively related to students' progress. Over the course of two academic years, we carried out an experiment in a single subject in the Computer Science degree in which students' emotional data and attendance were collected through "PresenceClick." Then, we analyzed the compiled data through a correlation study and a principal component analysis whose results demonstrate the consistency of the data, allowing the prediction models to be fed each academic year. Once the correctness and stability of data were verified, data mining techniques were applied and two models based on probability tables and decision trees were obtained. These models enable instructors and students to detect problems early and avoid failure.