Nonlinear time-series forecasting: A fuzzy-neural approach

Abstract

This paper presents a fuzzy-neural approach to the prediction of nonlinear time series. The underlying mechanism governing the time series, expressed as a set of IF-THEN rules, is discovered by a modified self-organizing counterpropagation network. The task of predicting the future is carried out by a fuzzy predictor on the basis of the extracted rules. We have applied the approach to three well-studied time series: sunspot, flour prices, and Mackey-Glass chaotic process. The results demonstrate that the approach is fairly effective and efficient in terms of relatively high prediction accuracy and fast learning speed. Comparative studies with other network approaches on these time series suggest that our approach can offer comparable or even better performances. One of the salient features of the approach is that, only a single learning epoch is needed, thereby providing a useful paradigm for some situations, where fast learning is critical.