A clustering-based characteristic model for unreliable Sensor Network data

Wireless Sensor Networks (WSNs) that are deployed outdoors suffer from rough environmental conditions. Moreover, cheap hardware or energy management techniques like undervolting might lead to inaccurate sensing results and, thus, unreliable data. Hence we propose a characteristic model of every sensor node´s data to derive the `normal´ behavior of sensors statistically. Beside massively reducing the total amount of sensed data, this representative model can be used to detect discordant values and redundancies between nodes. Theoretical considerations as well as a functionality test of a server implementation with real WSN nodes show the features of this approach.