A new fuzzy time series forecasting model based on clustering technique and normal fuzzy function

Time series forecasting is of great interest to managers and scientists because of the numerous benefits it offers. This study proposes three main improvements for forecasting to time series. First, we establish the percentage variation series between two consecutive times and use an automatic algorithm to divide it into clusters with a suitable number. This algorithm also determines the specific elements in each cluster. Second, a new fuzzy function with a normal type is built for each cluster. Finally, we develop the forecasting rule based on the previous two improvements. By combining these enhancements, we obtain an effective model for forecasting. The proposed model is presented step-by-step and executed rapidly using the MATLAB procedure. Compared to many models tested on the M3-Competition set with 3003 series and the M4-Competition set with 100,000 series, the proposed model obtains outstanding results. It also achieves competitive results when compared to existing models across several benchmarks and real series.