A forecasting model for time series based on improvements from fuzzy clustering problem

Abstract

This article proposes a new fuzzy time series model that can interpolate historical data, and forecast effectively for the future. It is combination of the improved steps from the existing models. There are problems to use the percentage variations of series between consecutive periods of time as a universal set, to divide the universal set into clusters by the automatic algorithm based on the similarity between elements, to determine the relationships between elements in the series and the divided clusters by the improved fuzzy cluster analysis algorithm, and to interpolate the historical data and to forecast for future by new principle. The proposed algorithm is performed quickly and efficiently by the established Matlab procedure. It is illustrated by an example, and tested for many other data sets, especially for 3003 series in M3-Competition data. Comparing to the existing models, the proposed model always gives the best result. We also apply the proposed model in forecasting the salty peak for a coastal province of Vietnam. Examples and application show the potential of the studied problem.