Unsupervised feature selection algorithms for wireless sensor networks

A wireless sensor network (WSN) is a distributed measurement system deployed over a geographical area to acquire physical information which, depending on the nature of the monitoring phenomenon, can be spatially correlated in space and time. Spatial correlation, to be intended here at different levels, can be exploited to reduce the communication bandwidth, implement articulated sensing and carry out energy saving policies. The paper aims at investigating unsupervised feature selection algorithms and how they can be used to exploit spatial correlation in WSNs. The interest is due to the fact that generation of a reduced set of features (i.e., aggregated data) has a positive effect on optimal energy management, hierarchical decision making and performance. Six algorithms have been critically discussed and contrasted both at theoretical and experimental levels.