Location-aware Affinity Propagation Clustering in Wireless Sensor Networks

Clustering has become a crucial operation in Wireless Sensor Networks (WSNs). Affinity Propagation (AP) is a relatively new clustering technique that has been shown to possess several advantages over long-standing algorithms such as K-means, particularly in terms of quality of clustering and multi-criteria support. However, the original AP algorithm is computationally intensive making it unsuitable for clustering in WSNs. A hierarchical decentralized variation of AP (Hi-WAP) has been recently proposed to reduce the processing cost of AP while minimizing the potentially negative effect of distribution (due to the lack of a global view) on clustering quality. In this paper, we explore the suitability of Hi-WAP for clustering in WSNs. We employ the level of distortion and the processing time as evaluation metrics. We propose an extension to Hi-WAP, termed LAP; Location-aware Affinity Propagation, where clustering is performed while being cognizant of nodes' locations. Simulation results reveal that LAP, in general, outperforms Hi-WAP. We further study the optimization of LAP parameter values with the objective of minimizing processing time while maintaining a desirable low level of distortion.