Incremental approach for detecting arbitrary and embedded cluster structures

Abstract

In this paper, we present a new incremental clustering approach (InDEC) capable of detecting arbitrary cluster structures. Cluster may contain embedded structures. Available methods do not address this important issue in the context of continuously growing databases. A density variation concept is used to detect embedded clusters that may occurs after successive updation of database. Unlike popular methods which use distance measure, we use a new affinity score to decide the proximity of a new object with the clusters. We use both synthetic and real datasets to evaluate the performance of our proposed method. Experimental result reveals that proposed method is effective in detecting arbitrary and embedded clusters in dynamic scenario.