Towards a Bayesian Student Model for Detecting Decimal Misconceptions

This paper describes the development and evaluation of a Bayesian network model of student misconceptions in the domain of decimals. The Bayesian model supports a remote adaptation service for an intelligent tutoring system within a project focused on adaptively presenting erroneous examples to students. We have evaluated the accuracy of the student model by comparing its predictions to the outcomes of the interactions of 255 students with the software. Students' logs were used for retrospective training of the Bayesian network parameters. The accuracy of the student model was evaluated from three different perspectives: its ability to predict the outcome of an individual student's answer, the correctness of the answer, and the presence of a particular misconception. The results show that the model is capable of producing predictions of high accuracy (up to 87%).