Weighted fuzzy time series forecasting based on improved fuzzy C-means clustering algorithm

Abstract

A novel method for fuzzy time series (FTS) forecasting is presented based on improved fuzzy C-means clustering algorithm (IFCM) and first-order difference. Traditional forecasting approaches have weighted the central values of fuzzy intervals corresponding to fuzzy sets, but the central values may not be accurate enough since the assumed membership functions may be different. To avoid the problem of even distribution, in this paper, we weight the cluster centers derived from IFCM that defines the initial cluster centers of traditional fuzzy C-means clustering algorithm (FCM). There are many unstable characteristics in the time series forecasting model. To eliminate the fluctuation tendency of unstable characteristics, the first-order difference is used as the smooth time sequence to observe. Our experimental results on Alabama University enrollments and Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) demonstrate that the effectiveness and superiority of the proposed forecasting approach, in this paper, which gets higher forecasting accuracy than state-of-the-art methods.