Expressions of Time Series in Natural Language with Global Trends and Local Features

- Defining of Fuzzy Intervals in the Temporal Axis Based on Tendency of Time Series

Abstract

We have many kinds of data of time series, which are understood via linguistic expressions by humans. We have proposed a method to express their global trend and local features in a natural language. In the previous research, however, linguistic expression does not explain some of time series, because all time series use fuzzy sets of the same definition for three fuzzy intervals in the temporal axis. When humans define some fuzzy intervals, they usually take the big turning point into account. In this paper, we propose a method to divide time series into some dusters and define fuzzy intervals based on the turning point by grouping similar tendency of adjacent dusters. We show that the global trends using this method explain time series more appropriately.