Time Series Modelling and Prediction using Fuzzy Trend Information

This paper presents a novel approach to modelling time series datasets using fuzzy trend information. A time series is described using natural linguistic terms such as rising more steeply and falling less steeply. These natural shape descriptors enable us to produce a glass box model of the series. The linguistic shape descriptors are represented in our system by a new feature called the trend fuzzy set. All trend fuzzy sets are derived from a window on the time series and as such describe the shape of the series within that window. Each window can have membership in any number of different trend fuzzy sets. Prediction using these trend fuzzy sets is performed using the Fril evidential logic rule. Examples of series prediction are shown using sine wave and sunspot time series data.