The Study of the Data Mining Oriented Method for the Symbolization of Time Series

Bin Li
2000-01-01
Abstract:In the process of data mining on time series, continuous time series needs to be transformed into discrete symbolic sequences. This paper proposes a simple but very efficient method for the symbolization of time series. This method has two features: First, it segments the time series by an approach called linear segmentation, which is well known for its shape expression and segmentation. Secondly it implements the online clustering analysis on the segmented time series with a fuzzy clustering algorithm using neural network. The measure of shape similarity is also proposed in this paper. Effectiveness of this method has been verified by the symbolization of sample financial time series.
What problem does this paper attempt to address?