Analysis And Prediction Of Student Emotions While Doing Programming Exercises

The modeling of student emotions has recently considerable interest in the field of intelligent tutoring systems. However, most approaches are applied in typical interaction models characterized by frequent communication or dialogue between the student and the tutoring model. In this paper, we analyze emotions while students are writing computer programs without any human or agent communication to induce displays of affect. We use a combination of features derived from typing logs, compilation logs, and a video of the students' face while solving coding exercises and determine how they can be used to predict affect. We find that combining pose-based, face-based, and log-based features can train models that predict affect with good accuracy above chance levels and that certain features are discriminative in this task.