Skip to Main Content
In order to improve the performance of traditional intrusion detection system (IDS) based on neural network, we design and implement an integrated model of IDS based on rough set and wavelet neural network (RWNN-IDS). This paper focuses on applying RWNN for attacks recognition. We first present a conditional information entropy based algorithm to select the smallest features set, which can ensure the correct classification. We then design a novel wavelet neural network(WNN), its hidden unit is a multi dimensional non-product wavelet-sigmoid basis function. A set of heuristic learning rules using dynamic hidden nodes method is presented for attacks classification. Constructing and training the WNN with tidy training set can improve generalization capability and classification accuracy. The experimental results show that this system outperforms traditional neural network based IDS in terms of detection speed and accuracy, computational cost, system adaptability. It also provides new ideas to improve performance of distributed IDS.