From students’ careers to recommendations: a process mining approach for next exam suggestion

In some University contexts, such as in the Italian University system, there is quite some flexibility on the order in which students can take the exams. While this flexibility can be advantageous, it may also influence students' performance and, in particular, their graduation time. In this work, we investigate the use of Educational Process Mining - an approach that integrates process mining, data mining, and learning analytics - to design a recommendation system that supports students in identifying the most appropriate exam(s) to take next. To this aim, we model the exams taken by students as Instance Graphs (IGs), which explicitly capture concurrency, intended as exams that students prepare simultaneously. The IGs derived from timely students, i.e., those who graduate within the expected timeframe, are then used to train a Graph Neural Network that learns their virtuous behaviors. The resulting model is then leveraged to provide datadriven recommendations based on students' current academic paths. Experiments were conducted on real data extracted from a Bachelor's degree program in engineering. The results highlight the good performance of the recommendation system, achieving an accuracy of 79.54%.