Wireless sensor network is simply a pool of self-directed devices organized into mutually connected network. Sensors are usually autonomous and spatially distributed within a certain area to monitor targeted physical and environmental conditions such as temperature, sound and pressure etc.In WSN, free frequency band and open architecture are used for supporting mission critical application in a hostile environment, these are highly prone to various security attacks such as wormhole attack.
A wormhole attack is considered to be a detrimental security threat for WSN. In WSN, known communication channel is used, so that the wormhole attack can be launched silently without compromising the security means.Wormhole attackers (node) are connected via virtual tunnel which can be established in many ways, such as out of bound hidden channel, packet encapsulation and high powered transmission. During this attack, malicious node recordspackets from one location in the network, and replays them in another location by sending through virtual tunnel to another malicious node.As depicted in fig-1, E1 and E2 are two wormhole end points, connected by dedicated link, can capture the packets from one location and replays it to another location.
Subsequently, this wormhole attack is so severe that it might destroy the network or hamper the usual operation of network by selective dropping of packets; Manipulation of traffic or modifying data packet without revealing their identity.
Therefore, detection of wormhole node is an essential task for ensuring the security of wireless sensor networks. Most of the existing countermeasure against wormhole attacks are based on distance between nodes , direction , and location abnormality  among claimed neighbour nodes. To gain certain level of accuracy, many existing schemes have been proposed to use sophisticated device such as directional antenna , GPS (for strict location) , or ultra sound for distance measurement . In fact, those special devices are too expensive for practical deployment. There are a few statistical or analytical based schemes to use hope count , node connectivity , or neighbourhood count  to detect wormhole nodes that do not need any special hardware. But those schemes are used with hardware based approaches. Furthermore, those methods caused significant network overhead, as its calculation performed at a central location of the network. However, most of the wormhole detection schemes are made to apprehend wormhole node in uniformly distributed sensor networks, but their performance in case of non-uniformly distributed network is in question.
In this paper, we are proposing a novel detection scheme based on artificial neural network (ANN) using number of neighbors. The proposed detection model, able to detect wormhole attack in non-uniform sensor distribution, does not need any special hardware. Here, we have introduced a mobile node, called as detector node (DN) that visits random location within the network area and collects neighborhood count. When DN moves into wormhole attacked zone, the collected number of neighbors are increased abruptly (uniform network scenario) or slightly (non-uniform network scenario) compare to non-affected zone. This abnormality is captured by DN as evidence of the presence of wormhole attack and gathered in a dataset. DN collects number of neighbors both in presence and absence of wormhole node. Dataset is used for Training and testing of neural network. After training phase, test data set are fed into neural network and based on output of the network, we decide the existence of wormhole attack in the network.
...(download the rest of the essay above)