A computational fuzzy time series forecasting model based on GEM-based discretization and hierarchical fuzzy logical rules

Abstract

Fuzzy time series forecasting models have been widely used for predictions in various domains, including the prediction of stock prices, project costs, academic enrollment, electric load demand, etc. Current studies in this field mainly focus on three issues: the discretization of real numbers, the expression of evolutionary rules generated from training data and the defuzzification of the forecasted fuzzy results. To automatically and intelligently determine the discretization intervals, this paper introduces a general entropy measuring (GEM) method into the partitioning process of the original time series. A computational algorithm is also designed to realize the auto-determination process for each subset. Then, an improved hierarchical architecture is employed to express the fuzzy logical evolutionary rules of the fuzzy time series. To compare the performance of the proposed model with that of other models, the commonly used Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) dataset is used as experimental data. The forecasting results are evaluated using the root mean squared error (RMSE). The comparison results demonstrate the superiority of the proposed model. Finally, the Shanghai Stock Exchange Composite Index (SHSECI) stock price datasets from 1991 to 2014 are collected and used to test the model’s applicability. The empirical results show that the proposed model can effectively handle large online datasets.