Synthetic Feature Transformation with RBF Neural Network to Improve the Intrusion Detection System Accuracy and Decrease Computational Costs

Abstract

With the rapidly growing and wide spread use of computer networks, the number of new attacks and malicious has grown widely. Intrusion Detection System can identify the attacks and protect the systems successfully. However, performance of IDS related to feature extraction and selection phases. In this paper, we proposed new feature transformation to overcome this weakness. For this aim, we combined LDA and PCA as feature transformation and RBF Neural Network as classifier. RBF Neural Net (RBF-NN) has a high speed in classification and low computational costs. Hence, the proposed method can be use in real time systems. Our results on KDDCUP99 shows our proposed method have better performance related to other feature transformation methods such as LDA, PCA, Kernel Discriminant Analysis (KDA) and Local Linear Embedding (LLE).