Hierarchical distributed data classification in wireless sensor networks

Wireless sensor networks promise an unprecedented opportunity to monitor physical environments via inexpensive wireless embedded devices. Given the sheer amount of sensed data, efficient classification of them becomes a critical task in many sensor network applications. The large scale and the stringent energy constraints of such networks however challenge the conventional classification techniques that demand enormous storage space and centralized computation. In this paper, we propose a novel decision-tree-based hierarchical distributed classification approach, in which local classifiers are built by individual sensors and merged along the routing path forming a spanning tree. The classifiers are iteratively enhanced by combining strategically generated pseudo data and new local data, eventually converging to a global classifier for the whole network. We also introduce some control factors to facilitate the effectiveness of our approach. Through extensive simulations, we study the impact of the introduced control factors, and demonstrate that our approach maintains high classification accuracy with very low storage and communication overhead. The approach also addresses a critical issue of heterogeneous data distribution among the sensors.