Modeling student pathways in a physics bachelor's degree program

TL;DR

This paper applies modern predictive machine learning models to analyze student pathways in a physics bachelor's program, revealing key factors influencing course-taking patterns and degree switching behaviors.

Contribution

Introduces a machine learning approach to predict student pathways in physics education research, highlighting the limited role of grades and demographics compared to course enrollment patterns.

Findings

Students switching to engineering often skip key physics courses.

Course performance and demographics have smaller predictive power.

Modern models outperform traditional statistical analyses.

Abstract

Physics education research has used quantitative modeling techniques to explore learning, affect, and other aspects of physics education. However, these studies have rarely examined the predictive output of the models, instead focusing on the inferences or causal relationships observed in various data sets. This research introduces a modern predictive modeling approach to the PER community using transcript data for students declaring physics majors at Michigan State University (MSU). Using a machine learning model, this analysis demonstrates that students who switch from a physics degree program to an engineering degree program do not take the third semester course in thermodynamics and modern physics, and may take engineering courses while registered as a physics major. Performance in introductory physics and calculus courses, measured by grade as well as a students' declared gender and ethnicity play a much smaller role relative to the other features included the model. These results are used to compare traditional statistical analysis to a more modern modeling approach.