Improved BP Neural Network And DS Evidence Fusion Theory In Intrusion Detection System

There are still a lot of defects in practical applications of intrusion detection technology, such as the high rate of false alarm and missing alarm, relatively low efficiency of detection and the shortage of the degree of intelligence. In order to improve the accuracy of detection and reduce the rate of false alarm and missing alarm, most of the research focus to select the appropriate data source and data attributes, improve existing detection algorithms,find the new detection algorithms and improve the architecture of intrusion detection system and so on.This paper designes a model of intrusion detection system and combine the BP neural network with D-S evidence theory on identifying intrusion in accordance with the main problem for IDS,but the BP neural network’s convergence slowly, falling into local minimum easily and so on, we use the genetic algorithm to optimize BP neural network’s weights.This paper realizes BP neural network algorithm and D-S evidence theory in VC++6.0, select the intrusion data sets of KDD CUP99[1],and preprocess and select features from data according to the need. Because the excessive feature of the datas used in the experiment and large amount of calculation, this paper selects Wang Zhen She and Fang Min’s feature based on principal component analysis feature extraction to reduce the dimension of the excessive feature.Constructing the BP neural network in accordance with the actual needs and learning the intrusion behaviors, then using D-S evidence theory fuses recognition results of BP neural network.Finally, compared with BP neural network alone,the recognition efficiency of the combination between BP neural network and D-S evidence theory is better.