RWTH Aachen University
In many applications, it is useful to detect the evolving pat-terns in a data stream, and be able to capture them accurately (e.g. detecting the purchasing trends of customers over time on an e-commerce website). Data stream mining is challenging because of harsh constraints due to the continuous arrival of huge amounts of data that prevent un-limited storage and processing in memory, and the lack of control over the data arrival pattern. In this paper, the authors present a new approach to discover the evolving dense clusters in a dynamic data stream by incrementally updating the cluster parameters using a method based on robust statistics.