Degeneracy in Student Modeling with Dynamic Bayesian Networks in Intelligent Edu-Games

This paper investigates the issue of degeneracy in student modeling with Dynamic Bayesian Network in Prime Climb, an intelligent educational game for practicing number factorization. We discuss that maximizing the common measure of predictive accuracy (i.e. end accuracy) of the student model may not necessarily ensure trusted assessment of learning in the student and that, it could result in implausible inferences about the student. An approach which bounds the parameters of the model has been applied to avoid the issue of degeneracy in the student model to a high extent without significantly diminishing the predictive accuracy of the student model.