A dynamic hierarchical incremental learning-based supervised clustering for data stream with considering concept drift

Clustering analysis is an important data mining method for data stream. Data stream clustering is a branch of clustering in which the patterns are processed in an ordered sequence. Data stream clustering faces various challenges due to high speed, high volume, evolutionary, unstable and unlimited nature of the data. Over time, the data confront with changes in space of input. This obstacle is called concept drift, where its investigation is of high importance. In this paper, a new method called dynamic clustering of data stream with considering concept drift is developed, which is an incremental supervised clustering algorithm. In the proposed algorithm, data stream is automatically clustered in a supervised manner, where the clusters whose values decrease over time are identified and then eliminated. Moreover, the generated clusters can be used to classify unlabeled data. Experimental results on 15 UCI datasets show that the proposed method outperforms the existing techniques.