Integrating Grade Prediction for Better Student Support in MIT’s Introductory Programming Course

Abstract

We focus on grade prediction in the context of the 6.0001/2 course by utilizing student data – including assignment, assessment, and participation scores/metrics and click data – from past iterations of the course. In doing so, we explore various machine learning algorithms to create expressive, accurate predictive models. We have created and integrated the predictive modeling tool into the current course site to allow course staff to monitor student grade trajectories while guiding and assisting struggling students. Staff are able to interface with this tool which allows them to see this grade prediction along with other useful attributes for any student enrolled in the course. Students, although not directly able to interface with the tool, can be alerted and offered assistance if their grade trajectory is predicted to be a failing grade or below a certain threshold in order to provide them with the necessary resources to succeed in the course. Finally, we analyze the grade predictions and compare them to the final grade outcomes to find trends and patterns with regards to how students with different trajectories through the semester adjust their behaviors in the course based on the marks they receive.