Cybersecurity is an ever-evolving field that demands more workers and a wider array of knowledge every year. As such, cybersecurity education remains essential — not just for professionals, but for developers and non-technical roles as well. Due to this, hands-on cybersecurity exercises, such as the ones in the eduRange platform, are increasingly important. EduRange aims to be a flexible, intuitive cybersecurity platform that allows instructors to tailor pre-existing scenarios to their classes’ needs. However, when students become stuck and annoyed, learning grinds to a halt. To combat this frustration, we want to create an automated hints system that can consistently identify struggling students. Such a hints system, however, requires a large quantity of data, which can be difficult to obtain through classroom testing.

As such, we explored creating synthetic data. We used a sample dataset and stored attempt accuracy in a three dimensional tensor with dimensions students, questions, and attempts. We then used tensor decomposition to fill in gaps in the dataset, a process called densification. Our primary objective was to optimize the tensor decomposition to obtain the most accurate possible densification. The results showed that to obtain the greatest accuracy, we should use rank-1 tensors and fill in logical extra data points. The results also implied that generic tensor decomposition may not be sufficient for boolean data, but a new path forwards could be using boolean tensor decomposition.