Strong α-cut and associated membership-based modeling for fuzzy time series forecasting

In this paper, a method is proposed to deal with factors affecting the fuzzy time series forecasting. A new fuzzification process is used by considering all the fuzzy sets with nonzero membership values corresponding to the data points. A strong alpha-cut based method is presented to select appropriate fuzzy logical relationships that carry importance in analyzing the trend of time series. Further, a unique defuzzification approach based on weights is proposed to get crisp variation. This obtained variation is finally converted to the forecasted value. The presented method is tested on the benchmark enrolment dataset of Alabama University and seven datasets of the Taiwan Capitalization Weighted Stock Index. On comparing the results, it is observed that the presented method performs better than the existing methods. Also, the statistical measures indicate the good forecasting results of the presented method.