Bayesian network analysis of computer science grade distributions

Time to completion is a major factor in determining the total cost of a college degree. In an effort to reduce the number of students taking more than four years to complete a degree, we propose the use of Bayesian networks to predict student grades, given past performance in prerequisite courses. This is an intuitive approach because the necessary structure of any Bayesian network must be a directed acyclic graph, which is also the case for prerequisite graphs. We demonstrate that building a Bayesian network directly from the prerequisite graph results in effective predictions, and demonstrate a few applications of the resulting network in areas of identifying struggling students and deciding upon which courses a department should allocate tutoring resources. We find that many of our observations agree with what has long been considered conventional wisdom in computer science education.