Applications of Machine Learning to Student Grade Prediction in Quantitative Business Courses

With ever growing pressure to increase student performance, as well as the need to support first generation students and those from underrepresented backgrounds, never before has there been a greater need to quickly identify students who are on track to perform poorly in a class. Traditionally, instructors have relied mainly on intuition and rough averaging of exam scores to predict what grade a student who is on track to receive. With the advent of machine learning and data mining, we have the unprecedented ability to leverage historical student performance data to make statistically accurate and justified predictions beyond an instructor's intuition. Here, we demonstrate the efficacy of three machine learning algorithms-Nave Bayes estimation, Knearest neighbors, and support vector machine-for use in predicting students' grades, and compare the results to a "simple average" approach of predicting student grades as a proxy for an instructor's prediction. We show that a support vector machine outperforms all other predictors, including the simple average. However, the error of the simple average predictor is not significantly greater than with the support vector machine, and with the increased computational cost of the support vector machine over the simple average predictor, it may be more economical and convenient to instead rely on instructor predictions or a simple average predictor than a more complex algorithm. Overall, while we show here that machine learning is promising for applications in predicting student performance outcomes, there is still much work to be done towards applying this technology on a broad scale.