Interpretable Methods for Early Prediction of Student Performance in Programming Courses

Early prediction of student grades is important for identifying students at-risk of failing or students not achieving their goals, and providing timely interventions. There is a need to develop methods that are both accurate and interpretable, to help teachers understand student behaviour and take appropriate actions. In this paper, we present a decision tree based approach for predicting students’ final performance in online computer programming courses for high school students, based on log data from the first part of the course. We define and extract suitable features characterising student behaviour and show that it is possible to build compact and accurate decision trees, achieving an overall accuracy of 76% and 82–83% at one-third and mid-course respectively. We provide insights about the important factors affecting student performance and how the rules can be used to improve the student learning outcomes.