Classification of Students' Conceptual Understanding in STEM Education using Their Visual Attention Distributions: A Comparison of Three Machine-Learning Approaches

摘要

Line-Graphs play a central role in STEM education, for instance, for the instruction of mathematical concepts or for analyzing measurement data. Consequently, they have been studied intensively in the past years. However, despite this wide and frequent use, little is known about students' visual strategy when solving line-graph problems. In this work, we study two example line-graph problems addressing the slope and the area concept, and apply three supervised machine-learning approaches to classify the students performance using visual attention distributions measured via remote eye tracking. The results show the dominance of a large-margin classifier at small training data sets above random decision forests and a feed-forward artificial neural network. However, we observe a sensitivity of the large-margin classifier towards the discriminatory power of used features which provides a guide for a selection of machine learning algorithms for the optimization of adaptive learning environments.