With the rapid adoption of software-based learning in classrooms, it is increasingly important to design more intelligent educational software, a goal of the emerging field of educational data mining. In this work, we analyze student activities from using a learning tool for engineers, Mechanix, in order to find trends that may be used to make the software a better tutor, combining its natural, sketch-based input with intelligent, experience-based feedback. We see a significant correlation between student performance and the amount of time they work on a problem before submitting; students who attempt to 'game” the system by submitting their results too often perform worse than those who work longer (p< 0.05). We also found significance in the number of times a student attempted a problem before moving on, with a strong correlation between being willing to switch among problems and better performance (p< 0.05). Overall, we find that student trends like these could be paired with machine learning techniques to make more intelligent educational tools.