Adaboost based parameterized methods for wireless sensor networks

Abstract

In the new distributed architecture, intrusion detection is one of the main requirements. In our research, two adaboost algorithms have been proposed. The very first procedure is a traditional online adaboost algorithm, where we make use of decision stumps. Decision stumps will be regarded as weak classifiers. In the following second procedure we make use of an enhanced online adaboost algorithm with online Gaussian Mixture Model (GMM) which will be referred as weak classifiers. Additionally we will be having distributed incursion detection framework, where local parameterized models for incursion detection are formed in every node by Adaboost procedures. Global detection models are also built up at every node by the combination of local parametric models by means of minor quantity of examples in the node. This arrangement is attained by a procedure constructed on the Particle Swarm Optimization (PSO) and also Support Vector Machines(SVM). Incursions will be detected using Global model in every node. Data collected on experimental results shows enhanced online adaboost process with GMM gives improved and high detection rate and reduced false alarms than the previous Adaboost processes. Our two algorithms outperform the current incursion detection procedures. It can be seen that our SVM and PSO based algorithms efficiently combines local models into global models at every node. The Global models in the node can identify and alarm incursions types which can be found in different nodes without sharing of samples of those incursion types.