A Unique Network-based Intrusion Detection Model

The genetic computing model represented by Artificial Immune System (AIS) is a big break in morden bionics science. Artificial Immune Network (AIN) algorithm which has been dramatically improved these years is an important part of AIS theory. Based on the discussion of idiotypic network which is the most important part of AIN, we proposed an improvement on one of its most typical model: AINet.During the research on AINet model, we found a big flaw of this algorithm: the inobservance of the density information of the training data set. Based on this disadvantage, we introduced AINDD: a new Artificial Immune Model which respected the density and depth information of the data set. This algorithm is then applied to a experiment on Data Clustering. The result showed that AINDD avoided the pitfall of AINet.Then we introduced this model into Network Intrusion Detection experiments. The KDD CUP 99 data set was discussed and network intrusion characteristic was present after that. The experimental results showed that when directly applied AINDD into the huge amounts of high dimensional data set the algorithm could consume too much memory and time. And it was not easy to find the best parameters for the algorithm under this circumstance. So the model needs to be improved to cope with the huge amounts of high dimensional data.We refered to the Artificial Tissue model and tried to abate the data size before the training phase. The result showed that the new model, AINDD with Artifical Tissue paradigm (AINDDT), was much faster then the former. Then we successfully find the paremeters for the new model during the adjusting experiments and achieve much higher detection rate and much lower false positive rate.