PME-NA

Addressing Imbalance: Evaluating Whether Instructional Activities Classifications Differ by Mathematics Content Domains

• In Event: Addressing Imbalance: Evaluating Whether Instructional Activities Classifications Differ by Mathematics Content Domains

Description

In this study, we investigate whether an imbalanced dataset with respect to mathematics content domains in video recordings of elementary mathematics lessons will yield differences in an artificial intelligence (AI) model’s performance to classify instructional activities occurring in the videos. Given that there was an over representation of lessons in the domain of Number and Operation, we anticipated the model's performance would be biased towards this domain. However, we found the AI model favored Data Analysis and Probability, which we confirmed at an aggregated and individual activity level analysis. As computer-vision human activity recognition networks performance are impacted by several factors (e.g., lighting conditions), we are unable to confirm yet if this difference is solely due to mathematical content domain. Nevertheless, the findings speak to the analysis that will be needed as more user facing AI models are taken up in mathematics teaching and learning and mathematics education research.

Authors

• James Drimalla, University of Virginia

• Jonathan Kyle Foster, University at Albany