Online K-means clustering with adaptive dual cost functions
Priti Sharma, Amit Sharma · 2017
Developments in internet technology has resulted in boom in volume and velocity of data being generated. Analysis of data requires clustering methods essentially. Online clustering deals with the problem of clustering the data as it arrives and restricts its storage. Rather only summaries of data like cluster assignment and representatives, number of clusters etc can only be stored. Conventional offline clustering methods cannot be extended easily for online purpose. This paper proposes how k-means can be adapted for online clustering through double objective functions. One is the local error estimate and the other is cost of opening new cluster. Cost of opening a cluster is exponential in the estimate of permissible cluster diameter. A regulation parameter is introduced to control the burst in number of output clusters.