The Mean-Variance Estimator Technique in Monitoring Applications Using Mobile Agents over Wireless Sensor Networks

We propose a reliable technique to detect behavior anomalies in monitoring critical infrastructures through resource constrained devices, for instance wireless sensor networks (WSNs). The study is specifically targeted to monitoring and alerting functionalities for homeland security, that typically enforce severe requirements to the detection process. Assuming the behavior of the characteristic operation indicators in a potentially large and complex infrastructure (such as buildings, bridges, nuclear power plants, aircrafts, etc.) to be bounded by design constraints, we can introduce a novel non-parametric detection technique that we denote as “MV-estimator-based” (where MV stands for sample mean and variance): the sample mean and the sample variance are computed from observations and behavior classification is performed by defining regions in the MV-estimator space instead of the observations space. It will be shown that the novel detection technique is able to provide better performance with respect to other approaches over resource constrained platforms such as WSN, and this will be substantiated by numerical results as well as by a detailed cost analysis. Moreover MVET operations into a clustered WSN are presented where MVET distributed functions are implemented by using mobile agents.