A Best-Match Forecasting Model for High-Order Fuzzy Time Series

An area of Fuzzy time series has attracted increasing interest in the past decade since Song and Chissom's pioneering work and Chen's milestone study. Various enhancements and generalizations have been subsequently proposed, including high-order fuzzy time series. One of the key steps in the Chen's framework is to derive fuzzy relationships existing in a fuzzy time series and to encode the relationships as IF-THEN production rules. A generic exact-match strategy is then applied to the forecasting process. However, the uncertainty and fuzziness characteristics inherent to the fuzzy relationships tend to be overlooked due to the nature of the matching strategies. This omission could lead to inferior forecasting outcomes, particularly in the case of high-order fuzzy time series. In this study, to overcome this shortcoming we propose a best-match strategy forecasting method based on the fuzzy similarity measure. The experiments concerning Taiwan Weighted Stock Index and Dow Jones Industrial Average are reported. We show the effectiveness of the model by running some comparative analysis using some models well-known in the literature.