Extreme learning machines for intrusion detection systems

Abstract

Information is a powerful tool that can be used as a competitive advantage to increase market shares, competitiveness and keep products up-to-date. Protecting the information is a difficult task; intrusion detection systems is one of the tools of great importance for the protection of computer network infrastructures. IDSs (Intrusion Detection Systems) are tools that help users and network administrators to keep safe from intruders and attacks of various natures. Machine learning techniques are one of the most popular techniques for IDSs proposed and investigated in the literature. This paper focuses on the use of ELM (Extreme Learning Machine) and OS-ELM (Online Sequential ELM) techniques applied to IDSs. Some features of these methods that motivate their use for building IDSs are: (i) easy assignment of parameters; (ii) good generalization; and (iii) fast and online training. The results show that the methods can be easily applied to a huge amount of data without a significant generalization loss. © 2012 Springer-Verlag.