An Interval-Type Autocorrelated Fuzzy Time-Series Model Used Other Fuzzification and Its Sequential Reconstruction

The Box-Jenkins model involves two interval-type autoregressive (IAR) models. In the IAR model proposed by Yabuuchi et al., fuzzified time-series data increased vagueness and sometimes produced an unnatural-forecast interval. Two approaches were previously proposed to solve this problem using a numerical example. The first is to change the target of fuzzification from time-series data to autocovariances. The second is to build the IAR model sequentially. The two approaches achieved the objective; however, in the model validation, widths of forecast intervals were very narrow. This paper considers the results of combining the two methods.