Mining Recent Approximate Frequent Items in Wireless Sensor Networks

Mining frequent items from sensory data is a major research problem in wireless sensor networks (WSNs) and it can be widely used in environmental monitoring. Conventional lossy counting algorithm can be applied to solve this problem in centralized manner. However, centralized algorithm brings severely data collision in WSNs, and results in inaccurate mining results. In this paper, we present D-FIMA, a distributed frequent items mining algorithm. D-FIMA, running at every sensor node, establishes items aggregation tree via forwarding mining request beforehand, and each node maintains local approximate frequent items. The root of the aggregation tree outputs the final global approximate frequent items. Theoretical analysis and the simulation results show that energy consumption of D-FIMA is much less than the centralized algorithm, and mining results of D-FIMA is more accurate than the centralized algorithm.