Impact of Packet Injection Models on Misbehaviour Detection Performance in Wireless Sensor Networks

Traffic in wireless sensor networks (WSN) is commonly created by sensor readings which can be modelled by various distributions as the occurrences of events triggers the injection of packets. The Poisson distribution is a common example for a widely used event distribution as many processes (for example the arrival of customers or the dissemination of parasits) are known to be Poisson distributed. The impact of such a packet injection model on sensor networks and especially on the performance of misbehaviour detection systems is therefore of interest. In this paper we investigate the impact of the Poisson model and the constant bit rate model on a misbehaviour detection system, namely an artificial immune system (AIS). We state the hypothesis that both models have no significant effect on the detection rate. We examine the influence of the two models on the detection performance and compare the results. We conclude that the differences between the two models show no statistically significant effects on the detection performance supporting our hypothesis. However, we observe that the AIS had a significantly smaller false positives rate for the Poisson model than for the CBR model.