Efficient hybrid rule pruning for intrusion detection using multi-dimensional probability distribution

Evolutionary algorithms for data mining have recently received increased attention due to their performance of the global search. Genetic Network Programming(GNP) has been proposed in recent years as one of the evolutionary algorithms and applied to data mining successfully, because of the prominent representation ability with the compact program derived from the directed graph structure and node reusability of GNP. Conventional GNP-based rule mining focused on binary-valued transaction data. Therefore, Fuzzy GNP based class association rule mining has been proposed to deal with the continuous-valued data types in the real network connection data. In this paper, firstly, many interesting rules are extracted by Fuzzy GNP-based hybrid class association rule mining from training data. Then, a post-processing method is used to prune class association rules. After that, a classifier is modeled based on the multi-dimensional probability distribution for testing data. Experiments on KDDCup 1999 data show the substantial improvement of the detection ability of the proposed method.