Data stream summarization methodology

Knowledge discovery from multidimensional data streams requires a method for effectively maintaining a historical "collection" of summaries of past windows such that historical stream queries can be answered. We present the Stream Summary Store method. In this method, 'similar' spatial summary objects that summarize points in different windows are approximated by one approximate spatio-temporal summary object. A stream summary store is a collection of approximate spatio-temporal summary objects that consume a certain space budget. We outline various policies for satisfying the space budget constraint during the summarization of the data stream, and focus on policies that utilize prominent data mining and compression techniques (Clustering and Signal Compression) to further reduce the space consumption of the summary store. An extensive experimental evaluation shows that our methodology for constructing approximate spatio-temporal summary stores coupled with well-known Clustering and Signal Compression techniques significantly outperforms existing methods for summarizing multidimensional data streams.