Mining Correlations Between Multi-streams Based on Haar Wavelet

Abstract: In the application with multiple data streams, the correlation between data streams is very significant. The main contributions of this paper included: (1) Introduces the concept of total ordering with filter and compression by wavelets to describe streams. (2) Proposes the equivalence model to evaluate correlations between streams, including three theorems about the equivalence between wavelet coefficients and original data about computing correlation. (3) Designs anti-noise algorithm with sliding windows to compute correlation measure. (4) Gives extensive experiments on real data, which show that the local correlations are hardly affected by data with noise in the long windows, and that new algorithm has well filter on the streams with noise in the environment of short size windows.