Bayesian network with Grey entropy data pre-processing for modeling students' learning status

Abstract

In this paper, our study aimed to use Grey entropy to help decide which attributes, so called items in educational assessment, should be eliminated to prevent the Bayesian network modeling process from over-fitting and to obtain better accuracy. Although Bayesian network is proving to be the best technology available for diagnosing students' learning status in educational assessment, in the process of constructing a Bayesian network, the criteria of selecting testing attributes such as items or tasks will influence the diagnosing accuracy. Experiment results indicats that the Bayesian network with Grey entropy data pre-processing obtains the better more than 10% in accuracy than the man-made Bayesian network.