The Behaviors Classification Based Malicious Nodes Detection In Wireless Sensor Networks

Wireless sensor network has features such as a lot of nodes, limited energy,limited node computing power and storage space, vulnerable to security attacks etc.The threats WSNs faced are not only the external attackers to network, the internalnodes of network may also be captured and become as the malicious nodes. Then, themalicious nodes may do worst damage to the whole network, if the malicious nodesof internal network are not detected timely. The wireless sensor networks will beparalyzed as the result. The traditional technology of encryption and authenticationcannot effectively be used to against network’s inner attacks, therefore effectiveintrusion detection technology has become the network’s second line of defence. Acooperation intrusion detection scheme based on combined weighted k nearestneighbor classification is designed and presented in this paper. Considering thenetwork energy efficiency at the same time, we joined IDS selection algorithm in theintrusion detection system. Finally, we simulationed for each index of system underthe simulation environment OMNET++4.1in different scenarios. Some goodsimulation of experimental data and results are analyzed.The important works of this thesis are as followed:1. After modeling the behavior attributes of wireless sensor network nodes (suchas packet loss rate, transferring rate, data packet forwarding delay, position matchinginformation), we had modeled some common network attacks, and established asuitable model of intrusion detection.2．Proposed malicious node detection mechanism based on combined weighted knearest neighbor method, Firstly collected data samples are randomly generatedattribute subset, then we run weighted k nearest neighbor classifier on each trainingsubset. Finally some better results of weighted k nearest neighbor classifier aregotten, after taking advantage of simple voting method.3．Due to the limited energy the sensor nodes carried, if intrusion detectionmodule (IDS) is deployed in each node, which will lead to decrease network lifeand increase the deployment cost of network. As far as possible in order to reducethe number of nodes which are needed to depoly intrusion detection module, but also to ensure all nodes in network can be coveraged by monitoring nodes, an IDSselection algorithm is designed and proposed in the new intrusion detectionscheme.The algorithm is deployed in the cluster nodes where IDSs are deployed also.The experimental verification results show that the new one not only optimizes theIDS nodes deployment, but also reaches an equilibrium network node energyconsumption and prolongs the whole wireless sensor network lifetime.