Title: Time series data analysis in multiple granularity levels
Author: Mehmet Sayal, Ming-Chien Shan
Hewlett-Packard Labs, 1501 Page Mill Road, Palo Alto, CA 94304 USA.
SAP Research, Palo Alto, CA 94304 USA
Abstract: A single-pass method for analysing time series data and extracting time-correlations (time-delayed relationships) among multiple time series data streams is described. The proposed method can detect and report time-delayed relationships among multiple time series data streams without having to transform the original data into another domain. Each time-correlation rule explains how the changes in the values of one set of time series data streams influence the values in another set of time series data streams. Those rules can be stored digitally and fed into various data analysis tools for further analysis. Performance experiments showed that the described method is 95% accurate and has a linear running time with respect to the amount of input data for pair-wise time series correlations.
Keywords: time series data; time domain analysis; correlation; convolution; granularity levels; granular computing; data analysis.
Int. J. of Granular Computing, Rough Sets and Intelligent Systems, 2009 Vol.1, No.1, pp.64 - 80
Available online: 24 Jun 2009