Accuracy enhancement for fuzzy time series using modified fuzzy relationship

Traditional fuzzy time series have more forecasting errors when clustering the universe of discourse is too low. This paper presents a new method using Cartesian product of a matrix to define the fuzzy relationship on the first order fuzzy time series. This technique can improve the accuracy and reduces a number of not-found relationships in historical time series compared to traditional method. In the experiment, it has two scenarios to consider; the forecasting enrollments of the University of Alabama and the forecasting foreign exchange rate from Euro to US$. The results represent the errors behaviors when changing values of various parameters and prove that proposed method yields better accuracy than the traditional method, especially in the low clustering of the universe of discourse and short historical time series.