Forecast combination with meta possibilistic fuzzy functions

Abstract

There are many methods to obtain accurate forecasts for time series data in the literature. It is imperative to find an appropriate method with the correct assumptions for a given data set and circumstances. However, the assumptions of existing individual methods rarely apply perfectly to data sets of real-life problems. Meta possibilistic fuzzy functions (MPFF) is introduced to overcome the limitations of individual methods by using meta fuzzy functions (MFF) in which the optimum function and weights for method aggregation are found. The possibilistic fuzzy c-means clustering algorithm is adapted in MFFs to mitigate the cost of misspecification of individual methods through weighted combination of methods in functions. The optimum effect sizes (weights) of the forecasting methods in the best function is determined from MPFFs. 9 real-world time series and a forecasting method are selected, and 1 real-world dataset and 13 different forecasting methods are determined for the experimental study of the proposed method. The results verified that the proposed approach achieves greater accuracy in terms of both mean absolute percentage error and root mean square error than existing forecasting methodology.