Intrusion Detection using NBHoeffding Rule based Decision Tree for Wireless Sensor Networks

Abstract

The objective of this paper is to build a practical intrusion detection system for wireless sensor networks which analyze the characteristics of traffic patterns and identify the intrusive activities in the network. It is to show that the choice of efficient and fast decision tree paradigm for intrusion detection with optimal features enhance the detection capability as well as saves energy, computation and memory of sensor networks. In addition, various rule based decision tree classifiers like Alternating Decision Tree, Decision Stump, J48, Logical Model Tree, Naive Bayes Tree and Fast Decision Tree learner have been compared with a family of Hoeffding rule based decision tree which shows better and fast detection capability. The evaluation of the enhanced feature space and the decision tree paradigm, on three different public dataset containing normal and anomalous data have been performed for various Hoeffding as well as other decision tree algorithms. With this approach it is proved that Hoeffding tree are best suited for online detection and handling of streaming sensor data with the efficient usage of memory in a resource constraint environment like sensor networks