Forecasting in fuzzy time series by an extension of simple exponential smoothing

Abstract

Fuzzy Time Series was introduced to improve the forecasting made by statistical methods in vague or imprecise data and in time series with few samples available. However, the integration of these concepts is a little explored area. In this paper we introduced a new forecast model composed by a preprocessing method and a predicting method. The pre-processing method is responsible for analyzing the data and defining a suitable structure of representation. The predicting method is based on the combination of fuzzy time series concepts with the simple exponential smoothing, a traditional statistical method for prediction. The experiments performed with the TAIEX index show that, besides obtaining better accuracy rates when compared with other methods available in the literature, the predictions made over the whole time series had the same behavior and trends than the real data.