Modeling Student Affective State Patterns during Self-Regulated Learning in Physics Playground

Shiming Kai
2019
This dissertation research focuses on investigating the incidence of student self-regulated learning behavior, and examines patterns in student affective states that accompany such self-regulated behavior. This dissertation leverages prediction models of student affective states in the Physics Playground educational game platform to identify common patterns in student affective states during use of self-regulated learning behavior. In Study 1, prediction models of student affective states are
more » ... veloped in the context of the educational game environment Physics Playground, using affective state observations and computer log data that had already been collected as part of a larger project. The performances of student affective state prediction models generated using a combination of the computer log and observational data are then compared against those of similar prediction models generated using video data collected at the same time. In Study 2, I apply these affective state prediction models to generate predictions of student affective states on a broader set of data collected from students participants playing Physics Playground. In parallel, I define aggregated behavioral features that represent the self-observation and strategic planning components of self-regulated learning. Affective state predictions are then mapped to playground level attempts that contain these self-regulated learning behavioral features, and sequential pattern mining is applied to the affective state predictions to identify the most common patterns in student emotions. Findings from Study 1 demonstrate that both video data and interaction log data can be used to predict student affective states with significant accuracy. Since the video data is a direct measure of student emotions, it shows better performance across most affective states. However, the interaction log data can be collected natively by Physics Playground and is able to be generalized more easily to other learning environments. Findings from Study 2 suggest that self-regul [...]
doi:10.7916/d8-z5be-xm41 fatcat:xinzfskwgnbchjuhoiqg7oymqu