Online fuzzy time series analysis based on entropy discretization and a Fast Fourier Transform

Abstract

Fuzzy time series analysis has been used successfully for forecasting in various domains including stock performance, academic enrollment, temperature, and traffic patterns. Research in this field has concentrated primarily on two issues: the reasonable partition of discourse, and defuzzification methods for discrete datasets. Both issues have a huge impact on the prediction performance of forecasting models. This paper integrates the entropy discretization technique with a Fast Fourier Transform (FFT) algorithm to develop a novel fuzzy time series forecasting model to resolve these issues. The Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) and Dow-Jones Industrial Average (DJIA) financial datasets were used to evaluate the model's performance. The results demonstrate that the presented model is a major improvement over previous fuzzy time series models produced by Chen (1996), Yu (2005), Chang et al. (2011), and Hsieh et al. (2011), and five other conventional time series models. The proposed model is implemented using the bootstrapping method, after which it incrementally updates its prediction capability. Results show that the proposed model's incremental learning mechanism allows it to effectively handle large online financial datasets.