Research and Implementation on Summary Data Generating Algorithm for Streaming Data Based on Haar Wavelet Transform

The study on streaming data is one of the hot topics among the database field recently. Unlike traditional data sets, stream data arrive continuously and they are fast changing, massive, possibly unpredictable. These characteristics of data stream determine that only approximate queries on them are proper. The key of approximate query is to construct a synopsis data structure far smaller than the size of the original data set with least reconstruction error. In this paper we present a novel greedy method , GreedyStream, to compute wavelet-based synopsis based on the particular analysis of wavelet theory. The synopsis produced by a conventional wavelet threshold approach have some significant drawbacks, such as the high variance in the quality of data approximation, the tendency for severe bias in favor of certain regions of the data. We present a one-pass wavelet threshold method to attempt to adapt the threshold based on what is happening to neighboring coefficients. Experiments on real data sets proved it more robust in accurate data reconstruction.