Network Intrusion Detection Using Flow Statistics

The increasing use of network data within every aspect of human life, ranging from genetic databases to credit card payments, urges for efficient methods for detecting any attempts (intrusions) to compromise sensitive information. The problem of detecting such network intrusions is challenging, since the regular or normal network patterns are permanently changing. This paper discusses a novel intrusion detection system based on using histograms of network parameters as features which are then fed into an extreme learning machine for classifying network flows. We evaluate and compare the proposed method with existing approaches using the ISCX-IDS 2012 benchmark dataset. The numerical experiments indicate that the proposed method outperforms existing approaches by achieving an average detection rate of up to 99% while suffering a misclassification rate of only 2 %.