Intrusion Detection in Dynamic Distributed Network Using PSO and SVM Machine Learning Algorithms

The number of computers connected to a network and the Internet is increasing with every day. This combined with the increase in networking speed has made intrusion detection a challenging process. System administrators today have to deal with larger number of systems connected to the networks that provide a variety of services. The challenge here is not only to be able to actively monitor all the systems but also to be able to react quickly to different events. Overall intrusion detection involves defense, detection, and importantly, reaction to the intrusion attempts. An intrusion detection system should try to address each of these issues to a high degree. So network security becomes more complex due to the arrival of large no. of new types of attack and lack of dynamic system to detect new type of attacks. In this paper we define the solution to frequently changing network environment. We propose Online Adaboost-based parameterized method. It contains two models, Local model and Global model. In the local model, online Gaussian mixture models (GMMs) and online Adaboost processes are used as weak classifiers. A global detection model is constructed by combining the local parametric model. This combination is achieved by using an algorithm based on support vector machines (SVM) and particle swarm optimization (PSO). This system is able to detect new types of attack when network environment changes. It gives high detection rate and low false alarm rate.