Embodied Learning With Virtual Reality: Gesture Clustering and Forecasting

• In Event: Understand Learning With Technology

Abstract

This study aims to model the spatial-temporal characteristics of student behaviors in virtual reality learning environments. The approach involves three steps: (1) K-Means clustering to determine a set of clusters for time series; (2) K-Nearest Neighbor to assign time series to clusters and validate gesture profiles; and (3) Linear Regression to forecast time series using cluster centroid as seed values. Analysis was conducted with 158 synthetic time series using quaternions to represent changes in orientation of a 3-dimensional virtual model of a dinosaur fossil over time. A total of six behavioral profiles were discovered, validated, and accurately forecasted from the time series. We discuss the implications for simulating embodied cognitive states in adaptive instructional systems.

Authors

• Matthew Orr, University of Utah

• Dany Aina, University of Utah

• Kirsten R. Butcher, University of Utah

• Eric G. Poitras, Dalhousie University