Abstract: To overcome the disadvantages of traditional decision tree algorithms, this paper proposed a relative decision entropy based decision tree algorithm DTRDE. First, we introduced the information entropy proposed by Shannon into rough set theory, defined a concept of relative decision entropy, and utilized the relative decision entropy to measure the significance of attributes. Second, in algorithm DTRDE, we adopted the relative decision entropy based significance of attributes and the dependency of attributes in rough sets to select splitting attributes. And we used the attribute reduction technology in rough sets to delete the redundant attributes,aiming to reduce the computation complexity of our algorithm. Finally, we applied the proposed algorithm to network intrusion detection. The experiments on KDI)Cup99 dataset demonstrate that DTRDE algorithm has higher detection rate than the traditional information entropy based algorithms,and its computational expense is simliar to those of the traditional methods.

Key words: Decision tree, Rough sets, Information entropy, Relative decision entropy, Significance of attributes, Intrution detection