Proactive reputation-based defense for MANETs using radial basis function neural networks

We have developed a proactive reputation-based defense system for Mobile ad hoc Networks (MANETs). In our work we assume the existence of nodal attributes which have the potential to affect the reputation score of a node at anytime. A radial basis function neural network (RBF-NN) is trained to learn the underlying mapping between the states of the various nodal attributes and the reputation score for the node at future times. Thus, the RBF-NN can be used to predict the reputation score of a particular node ahead of time, given only the current state of the node's attributes. Such a predictive system can result in lowering the reputation score of a node that is about to start malicious activity in advance of the actual attack. The RBF-NN predictors developed in this research to implement the proactive defense system resulted in an overall performance of 98.7% correct prediction with a 10-step predictor, and for comparison purposes, 98.1% with a 15-step predictor.