A Realtime Adaptive Trust Model Based on Artificial Neural Networks for Wireless Sensor Networks

Wireless sensor networks (WSNs) are vulnerable to security attacks due to the unbounded nature of the wireless medium, restricted node resources, and cooperative routing. Standard cryptography and authentication mechanisms help protect against external attacks, but a compromised node can easily bypass them. This work aims to protect WSNs against internal attacks, which are mostly launched from compromised nodes to disrupt the network's operation and/or reduce its performance. The trust and reputation management framework provides a routing cost function for selecting the best secure next hop. Tuning the trust weights is essential to cope with the constant changes in the network environment, such as the sensor nodes' behaviours and locations. To allow real-time operation, the proposed framework introduces an artificial neural network (ANN) in each sensor node that automatically adjusts the weights of the considered trust metrics according to the WSN state. A large dataset is generated to train and test the ANN using a multitude of simulated cases. A prototype is developed and tested using the J-Sim simulator to show the performance gain resulting from applying the adaptive trust model. The experimental results showed that the adaptive model has robust performance and has achieved an improved packet delivery ratio with reduced power consumption and reduced average packet loss. The results showed that when sensor nodes were static and malicious nodes were present, the average accuracy was 99.6%, while when they were in motion, it was 88.1%.