Analyzing Student Process Data in Game-Based Assessments with Bayesian Knowledge Tracing and Dynamic Bayesian Networks

Abstract

Digital game-based assessments generate student process data that is much more difficult to analyze than traditional assessments. The formative nature of game-based assessments permits students, through applying and practicing the targeted knowledge and skills during gameplay, to gain experiences, receive immediate feedback, and as a result, improve their skill mastery. Both Bayesian Knowledge Tracing and Dynamic Bayesian Networks are capable of updating students’ mastery levels based on their observed responses during the assessment. This paper investigates the use of these two models for analyzing student response process data from an interactive game-based assessment, Raging Skies. The game measures a set of knowledge and skill-based learner outcomes listed in a Canadian Provincial Grade 5 science program-of-study under the Weather Watch unit. To evaluate and compare the performance of Bayesian Knowledge Tracing and Dynamic Bayesian Networks, the classification consistency and accuracy are examined.