Modelling an Intrusion Detection system using ensemble approach based on voting to improve accuracy of base classifiers

Abstract

Security is always a major concern when people put their money or assets online. Since a network is an open road for data, unwanted users always try to make illegal use of these networks. Many important measures like authentication, authorization and use of firewalls were used in early stages to prevent such illegal access into systems. This work is aimed to study the various techniques and algorithms of an IDS and pick out some good techniques by doing a comparative study and then try to improve and overcome the design parameters of these techniques such that the overall design and functionality of an IDS is enhanced. In our proposed empirical/theoretical model the dataset which was reduced to a set of most prominent features based on experiment on three parameters. Intrusions are defined as activities or events that violate the confidentiality, integrity and availability of a computing system. Advancements in the field of artificial intelligence has given a different face to IDSs. In our model we have used three base classifiers. Our descriptive and comparative study helped us to short list the three algorithms. We have also taken a technique from the theory of induction through decision trees which is considered to be an ensemble in itself. Hence the overall accuracy of random trees is already better. This has helped to improve the overall accuracy of the proposed system. The final proposed model turned out to be more accurate and the time taken was also reduced by a large extent.