It's All in the Data - But What is It? Learning Analytics and Data Mining of Multimedia Physics Courses

Abstract

Before “Learning Analytics” became the buzzword that it is today, fueled by the advent of "big data" MOOCs and increased institutional attention to retention and time-to-degree, fine-grained transactions within courses were analyzed as a unique window into student learning: what resources do learners access in which order, how much time do they spend with each resource, to what degree do different resources contribute to success in formative assessment, what is the effect of online discussion forums, how many attempts does it take which learners to solve online homework, and what is the extend and effect of unproductive behavior such as procrastination, guessing, and copying? We will focus on virtual and blended physics courses and use data mining methods to extract various predictive signatures of eventual learner success on exams. We also use classical test theory and item response theory to estimate both quality parameters of online homework and latent learner traits such as ability and the propensity to copy or guess answers. Based on these findings, we will make recommendations for course design.