Predicting student success in cybersecurity exercises with a support vector classifier

In this paper we explore if we can detect whether students are struggling to complete simple hands-on cybersecurity exercises based on their command line history. These exercises are becoming more popular, especially with the increase in remote instruction. However, students may struggle for many reasons, including lack of some skills, confusion by what is being asked, or confusion about how the testbed works. Using a small collection of annotated log files from a sample exercise on DeterLab, we were able to generate three features and construct a support vector classifier to predict with 80% accuracy if students would complete the remaining parts of the exercise. Our work could be applied to early detection of students who likely will have difficulty completing the exercise, and offer them hints to boost engagement and learning.