A Fuzzy-Grey Model for Non-stationary Time Series Prediction

In time series prediction, historical data are used as the basis of estimating future outcomes. Many methods including statistical predictive models and artificial intelligence (AI) based models have been proposed for time series prediction. When dealing with limited information, researchers tend to seek for AI-based approaches as statistical models require large samples to determine the underlying distribution. This paper introduces a novel approach using fuzzy interpolation in constructing new data points adaptively within the range of known data in the grey prediction model. Denoted as fuzzy-grey prediction models (FGPM), the proposed model can improve the prediction accuracy of conventional grey models in the application of non-stationary time series prediction. The proposed model was tested on a practical data set derived from Taiwan Stock Exchange Capitalization Weight Stock Index (TAIEX). Experimental results showed that the proposed FGPM has the ability of fitting non-stationary time series accurately and outperforms some existing methods.