Dynamically evolving clustering for data streams

In this paper, a new online evolving clustering approach for streaming data is proposed, named Dynamically Evolving Clustering method. The clustering approach attempts to meet the following three key requirements of data stream clustering: (i) fast and memory efficient (ii) adaptive (iii) robust to noise. The proposed clustering approach processes one sample at a time and makes necessary changes to the model and then forgets the processed sample. This feature naturally makes it adaptive to changes in the data pattern. The clustering method considers both distance and weight before generating new clusters. This avoids generation of large number of clusters. Further, to capture the dynamics of the data stream, the weight uses an exponential decay model. Since in data streaming environment, a low density cluster can be outlier points or seed of actual cluster, DEC applies a strategy that enables detecting and removing only those low density clusters that are real outliers. To evaluate the performance of the proposed clustering approach, experiments were conducted using benchmark dataset. The results show that the Dynamically Evolving Clustering approach can separate the data well which are evolving in nature