Automatic Determination of K in Distributed K-Means Clustering
Dinesh Kumar Kotary, Satyasai Jagannath Nanda · Procedia Computer Science · 2019
Traditional K-Means based distributed data clustering require number of clusters as input which is difficult to obtain in case of a real life application like wireless sensor network. To mitigate this issue here an Automatic Distributed K-Means (ADK-Means) algorithm is proposed. In this algorithm cluster assignment is carried out with point symmetry based distance instead of Euclidean distance, to effectively detect arbitrary shaped clusters. The performance of proposed algorithm is demonstrated on four synthetic and one practical dataset. Three cluster quality evaluation methods have been employed and comparative analysis is done with existing distributed K-Means. The improved performance is up to 12.6% in terms of Silhouette index reported as compared to existing approach.