A Distributed Neighbourhood DBSCAN Algorithm for Effective Data Clustering in Wireless Sensor Networks

Conventional K-Means based distributed data clustering has limitation of detecting arbitrary shape clusters and requires number of clusters a priori. To alleviate these issues in this paper, a Distributed Neighborhood DBSCAN (DN-DBSCAN) algorithm is introduced which mutually exchanges data between neighbor nodes to perform partitioning of collected sensor data. The algorithm shares selected core points (obtained after local DBSCAN at each node) among the neighboring nodes on which DBSCAN is again allowed to run which leads to the formation of universal clusters. Observing the universal clustering patterns each sensor node adjusts its local clusters via cluster relabeling. The simulation study of proposed method is carried out on an artificial dataset and two practical case studies: Intel Lab dataset and Lower Gordon Snow Pole transect dataset. The proposed approach supersedes the existing K-Means based distributed clustering approach considering accuracy and computational time.