2015 International Symposium of Computer Application and Information Technology (IsCait 2015)

294 / 1971-01-01 00:00:00

Mmifs And Srm-elm Based Rrbf Neural Network For Anomaly Detection

5798,5799,5800,5801,5802

This paper presents an anomaly detection algorithm. Firstly, using Modified Mutual information-based Feature Selection algorithm (MMIFS) to reduce the feature dimension of network traffic based on mutual information entropy, and selecting the main strong relevant feature with the attack flow as the input of the regularization Radial basis function (RRBF) neural network, then using the idea of structural risk minimization (SRM), establish SRM-ELM learning algorithm to train the neural network. The algorithm uses different optimal feature subset for different types of detection, while avoiding the defection of falling into local optimum easily in traditional neural network. Simulation results show the convergence speed and detection accuracy of RRBF are better than QWNN and PLSSVM.