A Novel Approach to Handle Forecasting Problems Based on Moving Average Two-Factor Fuzzy Time Series

Abstract

In this paper, we present a novel approach to handling forecasting problems based on moving average in two-factor fuzzy time series. The proposed method defines a new technique to partition the universe of discourse into number of intervals based on the number of observations available in the historical time series data. Partition of interval depends on the transformed moving average time series data rather than actual time series data sets. Further, triangular fuzzy set is defined for transformed moving average data set and obtained membership grades of each moving average datum to their corresponding triangular fuzzy sets. Also, variation data set is calculated from transformed moving average data sets to define second factor data set. Further, frequency occurrence of fuzzy logical relationships is used in defuzzification process. The proposed method of moving average forecasting is verified and certified with three different fuzzy time series models. The robustness of proposed method is implemented in forecasting of Bombay Stock Exchange (BSE) Sensex historical data and compared in terms of different statistical error which indicates that the proposed method can provide more accurate forecasted values over with existing fuzzy time series models.