Feature Selection for Wireless Sensor Networks

In a wireless sensor network, appropriate use of resources can be achieved by choosing the most relevant sensors that are important to the current context. By applying feature selection to determine the optimal sensor locations, the number of sensors can be reduced and great savings (in term of power, hardware, and transmission channel) can be achieved without degrading the decision process. This paper presents the use of BFFS (Bayesian Approach for Feature Selection), a filter based feature selection method, for optimum sensor location selection. The virtue of the method is that the selection of features is purely based on the data distribution and thus is unbiased towards a specific model. The strength of this framework is demonstrated by experiments on activity recognition with the use of a Self-Organising-Map.