A refined weighted method for forecasting based on type 2 fuzzy time series

Abstract

In this paper, we proposed a method for type 2 fuzzy time series forecasting which is an extension of type 1 fuzzy time series model to enhance the accuracy in forecasts. The proposed method uses frequency distribution approach to define the appropriate length of intervals. High and low observations are used to define type 2 fuzzy time series and different fuzzy logical relationship groups (FLRGs) have been obtained for both high and low observations. Further, weight function are defined with the help of FLRGs to compute forecasted outputs by a simple arithmetic mean rather than complicated union and intersection operator of type 2 fuzzy sets. The proposed method has been applied for forecasting university enrollments and crop (wheat) production. It is shown that the proposed method has higher accuracy in terms of mean absolute percent error and root-mean-square error (RMSE) as compared to the other fuzzy time series methods.