Time Series as a Point - A Novel Approach for Time Series Cluster Visualization

摘要

Temporal data mining is concerned with the analysis of temporal data and finding temporal patterns, regularities, trends, clusters in sets of temporal data. Wavelet transform provides a means to analyze a temporal data at multiple resolutions. In this paper we propose a methodology for representing a time series as histograms at different resolutions using wavelet transform. Then we fit a regression line on the cumulative histogram and express the line as a point in the Hough space. Thus we are able to express an entire time series as a single point. So we propose a method for visualizing the time series clusters in a scatter space as well as multiple resolutions. The technique is illustrated with a sample set consisting of 24 short segment time series.