Evaluation of students' source code submissions using Machine Learning

In recent years, machine learning is experiencing a new golden age with the implementation of such methods in a wide variety of fields of study. Among them, the use of such techniques for source code analysis is gaining ground in tasks such as code optimization, code suggestion and bug detection. This project is intended to be an introductory study for the application of these techniques in academic and didactic fields. In the course of this research we tried to analyze the source code developed by the students of the computer science course at the Ecole Polytechnique de Louvain. In particular, we tried to predict the outcome of the final exam taken by a student starting from the source code of the exercises carried out during the semester. In addition, a further analysis was conducted to understand whether the source code alone is sufficient to estimate the quality of an exercise. Finally, a study was made on the interpretability of the results obtained using the binary classifiers created. The analyses undertaken did not always lead to the desired results, but can be seen as a first step in a field of study that deserves more attention.