Structure searching for Bayes networks by using variable importance indices from CART

Abstract

We usually handle large data set with more than 100 variables in social statistical analysis or biological analysis. However, it is often impossible to handle 100 variables at once without the super-computer. Although we use the super-computer, it takes long time to compute and build its model at once. In this case we can choose a tree-structured approach as a solution, The CART provides users with many options and useful results in detail. We have to examine variable importance which the CART offers. After we select relatively important variables which reflects the contribution each variable makes in classifying or predicting the target variable, we can make a new grouping algorithm. Finally, we get more information by comparing the true model which is given in the form of a Bayes network with a new model.