Using Index Structures for Anytime Stream Mining

Stream data mining has gained a lot of attention over the last years due to an abundance of streaming data in professional as well as personal applications. Solutions have been proposed for many mining tasks such as clustering, classification, frequent item set mining and aggregation. Stream mining is especially challenging due to the large (usually endless) amount of data and the time constraints posed by the stream’s arrival rate. We recently presented an indexbased solution for anytime stream classification that handles both large amounts of data and arbitrary arrival times. In this paper we present our ongoing work, wherein we investigate bulk loading strategies to improve the classification accuracy w.r.t. anytime constraints. We show promising results and discuss future challenges related to index-based classification on data streams. Furthermore we discuss extensions of our technique to other data mining tasks.