Abstract:

In this study, we identify predictors of student performance (both good and poor) in an algorithms course, due to its integral nature within the computer science (CS) discipline. Typically, features of race and/or gender are considered in such studies. However, we explore these combined with other features that can serve as predictors, specifically:
1) transfer status,
2) GPA of a prerequisite discrete math course, and
3) the semester span between discrete math and algorithms.

The goal of the study is to use these additional features in statistical regression models to identify if they are significant predictors of academic performance exist for the algorithms course. The 12-year dataset used includes approximately 49,000 CS students from a US based, R1, diverse university with over 55% of undergraduate students being non–white, over 58% being female, and 30% first generation students receiving Pell grants. Through our investigation, we achieve our goal by finding that additional significant predictors of performance in an algorithms course do exist. We show that all features considered have impact on performance, however, we see that grades in the prerequisite course as the most dominant predictor, even more so than race and gender. Using these findings, one can design and deliver qualitative measurements that support mixed methods research and offer interventions targeting the subgroups as defined with combined co–features. Our approach can be used to analyze a combination of predictors for other CS courses, or it can be generalized for use in other disciplines of study.

Keywords:

Educational data mining, performance predictors.