Clustering Large Dynamic Datasets Using Exemplar Points

In this paper we present a method to cluster large datasets that change over time using incremental learning techniques. The approach is based on the dynamic representation of clusters that involves the use of two sets of representative points which are used to capture both the current shape of the cluster as well as the trend and type of change occuring in the data. The processing is done in an incremental point by point fashion and combines both data prediction and past history analysis to classify the unlabeled data. We present the results obtained using several datasets and compare the performance with the well known clustering algorithm CURE.