Signature extraction for event forecasting in wireless sensor networks

Motivated by earlier work on adaptive event forecasting, this paper proposes a procedural event signature extraction method for wireless sensor networks, and a probabilistic approach to model non i.i.d. (independent and identically distributed) aperiodic traffic, which is then used to demonstrate the effectiveness of the proposed signature extraction method in support of reliable event forecasting. Since the quality of forecasts is declining as the redundancy, the noise, and the size of the TSS database is increasing, it is imperative to extract the event signatures from the noisy and mixed event sequences. By discarding the irrelevant events from the database, we significantly increase the quality of future forecasts. The proposed method is able to provide the user (on demand) with a human readable form of event signatures (in contrast to black-box modeling techniques), which might be of great help in understanding the events that took place in the monitored environment. We evaluate the proposed extraction method by means of simulations and investigate its parameter sensitivity as well.