Preprocessing in Fuzzy Time Series to Improve the Forecasting Accuracy

Abstract

The preprocessing in fuzzy time series has an important role to improve the forecast accuracy. The definitions of domain, number of linguistic terms and of the membership function to each fuzzy set, has direct influence in the forecast results. Thus, this paper has the focus on definition of these parameters, before of performing the prediction. The experimental results in enrollments time series show that, when the forecast is performed after proposed preprocessing, the accuracy rate is improved.

A novel intuitionistic fuzzy time series prediction model with cascaded structure for financial time series

“…For the highorder prediction models, in some studies, entropy-based and other methodologies have been chosen to specify the interval lengths Chen, 1996;Liu, 2007;Zhao & Yang, 2009). Especially in recent years, fuzzy C-means (FCM) (Bezdek, 1981) and some other clustering techniques have been utilized to be able to realize fuzzification in a more systematic way (Aladag et al, 2012;Cheng et al, 2016;Cheng & Li, 2012;Chen et al, 2012;Dos Santos & De Arruda Camargo, 2013;Eǧrioǧlu, 2012;Li & Cheng, 2010;Wei et al, 2014;Yolcu et al, 2013). While in early studies fuzzy logic relation matrix has been used to determine the fuzzy relations (Song & Chissom, 1993a;Song & Chissom, 1993b;Song & Chissom, 1993c)subsequently, the use of transition matrices has been preferred (Sullivan & Woodall, 1994) After a while, ANNs became popular and attractive as fuzzy relation determination tool in the FTS modelling process (Aladag et al, 2009;Aladag et al, 2010;Egrioglu, 2014;Egrioglu et al, 2009a;Egrioglu et al, 2009b;Lee & Hong, 2015;Wang & Xiong, 2014;Wei et al, 2014;Yu & Huarng, 2008).…”