Using Classification and Clustering to Predict and Understand Student Behavior in an Innovation-Based Learning Course

Abstract

One of the Grand Challenges for Engineering is advancing personalized learning, but challenges remain to identify and understand potential student pathways. This is especially difficult in complex, open-ended learning environments such as innovation-based learning courses. Student data from an iteration of an innovation-based learning course were analyzed using two educational data mining techniques: classification and clustering. Classification was used to predict student success in the course by creating a model that was both interpretable and robust (accuracy over 0.8 and ROC AUC of over 0.95). Clustering grouped student behavior into four main categories: Innovators, Learners, Surveyors, and Surface Level. Furthermore, noteworthy variables from each model were extracted to discover what factors were most likely to lead to course success. The work presented contributes to gaining a better understanding of how engineering students innovate and brings us closer to solving the Grand Challenge of advancing personalized learning.