A Coalesce of SNE-Wavelet-SVM Technique for Network Intrusion Detection

Abstract

Recognizing intrusions quickly and precisely is vital to the proficient operation of computer networks. Precisely describing critical classes of intrusions extraordinarily encourages their recognizable proof; be that as it may, the nuances and complexities of anomalous activities can without much of a stretch complicate the procedure. Due to the inherent capability of the signal processing to discover the novel and obscure attacks, they have been pretty popular for Network Intrusion Detection, and the nearness of the self-comparability in the system activity propels the appropriateness for the application Wavelets. In this work we first subject the network data to dimension reduction using Stochastic Neighbor Embedding (SNE) and then preform the wavelet decomposition of the data. The classification results of the pre-processed data using Gaussian SVM over different bandwidths uphold the claim that the proposed system has appreciably improved detection coverage for all the attack groups and the normal data as well, and at the same time minimized the false alarms. (Coiflets), Biorthogonal wavelets, Harmonic wavelets, Legendre wavelets, M-band wavelets and Composite wavelets.