Intrusion detection analysis by integrating roulette wheel and pseudo-random into back propagation networks

Intrusion detection is a critical component of network security; detection schemes fundamentally use the observed characteristics of network packets as a basis for such determinations. In this study, a cluster center distance method is applied to classify packet type. The cluster center is determined using characteristics of a portion of selected packet data samples prior to detecting. Meanwhile, a well-known back-propagation neural network combined with the roulette wheel selection method and pseudo-random rule are combined with back propagation network (BPN) to determine the intrusion packet type. KDDCUP99 data sets were used as the evaluation packet samples of this experiment. Simulation results demonstrate that roulette wheel selection combined with BPN scheme provides higher detection rate for DoS and R2Lattack packets; BPN with pseudo-random rule can yield higher detection rate for U2R attack packets.