Intrusion Detection Techniques Based on Improved Intuitionistic Fuzzy Neural Networks

Abstract

At present, the issue of intrusion detection must be a hot point to all over the computer security area. In this paper, two novel intrusion detection techniques have been proposed. First, unlike the current existent detection methods, this paper combines the theories of both intuitionistic fuzzy sets (IFS) and artificial neural networks (ANN) together, which lead to much fewer iteration numbers, higher detection rates and sufficient stability. Experimental results show that the first method proposed in this paper is promising and has obvious superiorities over other current typical ones. Then we address another novel technique based on non-subsampled Shearlet transform (NSST) domain artificial neural networks (ANN) to solve those problems, including employing multi-scale geometry analysis (MGA) of NSST and the train characteristics of ANN together. Experimental results indicate that, compared with other existing conventional intrusion detection tools, the second proposed method is superior to other current popular ones in both aspects of iteration numbers and convergence rates. Therefore, the anomaly detection process is performed through processing the signals representing the metrics. Experimental results indicate that the third proposed technique is effective and promising.