Use of Professor Comments in Predicting Student Success

Abstract

During their studies students receive written notes and comments from their professors assessing their grades, attitudes, qualities, and lacuna. These characteristics reflect a more subjective approach as opposed to the typical grading system. This paper, through topic modelling and word vectorization approaches, uses textual data to predict at-risk students in their first year of university studies with a Random Forest model. First, we introduce the used methods and analyze the corpus at hand. Then we vectorize the data (by Latent Dirichlet Allocation and other vectorizing methods) to categorize it and use it in the classifier. We then propose adding a dynamic element to the prediction through linear regression when using our data as a time series. Finally, we will review the prediction accuracy and feature importance to assert if these professor comments do indeed reflect the student’s scholar capacities. After comparing with the raw numerical grade data, we have better or as-good-as results by using our augmented textual data.