Automated network feature weighting-based intrusion detection systems

A common problem for network intrusion detection systems is that there are many available features describing network traffic and feature values are highly irregular with burst nature. Some values such as octets transferred range several orders of magnitudes, from several bytes to million bytes. The role of network features depends on which pattern to be detected: normal or intrusive one. Intrusion detection rates would be better if we know which network features are more important for a particular pattern. We therefore propose an automated feature weighting method for network intrusion detection based on a fuzzy subspace approach. Experimental results show that the proposed weighting method can improve the detection rates.