Evolutionary Fuzzy Relational Modeling for Fuzzy Time Series Forecasting

The use of fuzzy time series has attracted considerable attention in studies that aim to make forecasts using uncertain information. However, most of the related studies do not use a learning mechanism to extract valuable information from historical data. In this study, we propose an evolutionary fuzzy forecasting model, in which a learning technique for a fuzzy relation matrix is designed to fit the historical data. Taking into consideration the causal relationships among the linguistic terms that are missing in many existing fuzzy time series forecasting models, this method can naturally smooth the defuzzification process, thus obtaining better results than many other fuzzy time series forecasting models, which tend to produce stepwise outcomes. The experimental results with two real datasets and four indicators show that the proposed model achieves a significant improvement in forecasting accuracy compared to earlier models.