The main purpose of the research is to build an adaptive diagnostic test and remedial instruction system. We use Bayesian networks to make models of mathematical concepts and identify bugs and sub-skills in "prism, pyramid, cylinder and cone" unit. Besides、we try to combine multiple Bayesian networks to get better classification results than single Bayesian networks. The results show that the fusion method "sub-structure fusion", with dynamic cut-point selection can improve the classification accuracy. At last, we experiment on sixth grade elementary school student with the system. The results show that using Bayesian networks to diagnose the existence of bugs and sub-skills in individual students can get good performance. Also, students have made great progress in their mathematics studies.
