A case study of early performance prediction and intervention in a computer science course

Abstract

This study presents the results of a course intervention performed in a large upper-division undergraduate computer science class designed to offer additional resources to students that were identified as at-risk of low performance after completing graded assessments during the first two weeks of the semester. The course uses Python as the required programming language, however not every student that takes the class has prior experience with Python. The disparity in programming skills can greatly affect the overall student's experience in the classroom and potentially their overall course performance. We used data from the first two quizzes and one homework assignment from previous semesters to train a model using machine learning algorithms in order to predict students that were at risk of lower performance. At the end of week 2, students identified as at-risk received an invitation to join a 6-week course, which was created to give students an additional opportunity to work on programming activities using Python. The tasks involved real world examples, designed in a structured way to allow students to complete the solution on their own, without a lot of guidance from the instructors. Focus groups were conducted to capture student perceptions of the course.