Adaptive Trust Model for Secure Geographic Routing in Wireless Sensor Networks P. Raghu Vamsi and Krishna Kant Department of Computer Science and Engineering Jaypee Institute of Information Technology, Noida, India. prvonline@yahoo.co.in, k.kant@yahoo.co.in Abstract—Adaptive Trust Model (ATM) for secure geographic routing has been presented in this paper. Unlike conventional weight based trust models, ATM adjusts the weights associated with the network activities. In this model, direct trust has been considered to restrict the reputation based attacks. Due to the flexibility of weights adjustment, ATM dynamically identifies malicious nodes and directs the route towards trustworthy nodes. This ATM has been integrated into Greedy Perimeter Stateless Routing (GPSR) protocol. Simulation results using the network simulator ns-2 have shown that GPSR with ATM is robust in detecting malicious nodes. Keywords—Trust, Adaptive trust model, security attacks, secure routing. I. I NTRODUCTION Wireless Sensor Networks (WSNs) are a special class of wireless networks in which the network is composed by tiny and low-cost sensor nodes (SNs) having limited resources in-terms of processing, memory and energy. Applications of WSNs include temperature and humidity monitoring, pollu- tion monitoring in environmental applications, supply chain management applications, body area networks, pressure and speed monitoring in automotive, pungent gas or chemical detection in industries, target detection in military etc. [1]. In general, these networks are deployed in open, unattended and hostile environments to monitor and report the events. Due to the openness of SNs in WSNs, they are subject to eavesdropping, physical tampering, etc., which leads to various security attacks. Research studies in Mobile Adhoc Networks (MANETs), which is regarded as the predecessor of WSNs, have shown that packet delivery percentage can substantially decrease when malicious activities are present in the network. Cryptography methods can be utilized to mitigate security attacks posed by malicious nodes in the network and to provide authentication and secrecy. However, these methods require thousands of multiplication and addition operations to implement a single security operation. In addition, they are not efficient in identifying security attacks posed by an adversary from outside of the network. In other words, cryptography can aid communication security rather than routing security. Due to this reason, cryptography methods are not suited for routing in resource constrained sensor networks. For routing security a human behavioral pattern called trust has been widely used by the researchers. Trust is the measure of belief about the behavior of other entities (or nodes). Each node in the network assesses the trust of the neighboring nodes based on cooperation and coordination received in the network activities such as packet forwarding, maintaining packet integrity etc. [2]. The trust level will be incremented by one unit for every positive experience and decremented by one unit for every negative experience. In addition, special value called weight is assigned to every network activity. Trust models compute the total trust of a node by combining weights and trust opinions. These trust models are integrated with routing protocols to make efficient routing decisions. Among all routing protocols, geographic routing offers guaranteed packet delivery in a dense network. These routing protocols perform routing based on the location information. A node forwards the packet to a node which is situated nearer to the destination. Whenever malicious nodes are present in the network, they modify or tamper the packet integrity so that a benign node drops the packets as invalid. To mitigate such malicious activities, numerous researchers have devel- oped trust models to facilitate routing protocols. Weighted statistical model to identify malicious nodes is presented in [3]. Reputation based trust models that compute total trust by combining direct and indirect trust opinions is proposed in [4-8]. Trust model based on behavior of neighboring nodes has suggested in [9][10]. In addition, heuristic frameworks for trust management have been evolved in recent years [11]. Most of these models apply weights to compute total trust value. However, these weights are heuristic assignments. However, there is no proper mechanism to fix weight values. In human behavior pattern, trust on a person will be formed based on the number of conversations made and increases with the number of positive experiences with him. This implies that in order to gain trust and to provide weightage to a person there should be adequate number of interactions. With this analogy, an Adaptive Trust Model (ATM) has been presented in this paper. In this model, a node will dynamically adjust the weights based on the experiences gained and interactions held with its neighboring nodes. Weights are associated with expectations of various network activities to compute total trust value. Finally, during routing process, the packets are routed to nodes having the highest trust value. After explaining preliminaries and related work in Section II, network model and assumptions for ATM are presented in Section III. Section IV describes the ATM. Simulation study to validate the ATM is presented in Section V. Finally, Section VI concludes the paper. 978-1-4799-5173-4/14/$31.00 ©2014 IEEE 394