A non-parametricmodel for fuzzy forecasting time series data

The time series analysis is mainly aimed at establishing a fuzzy prediction model based on a set of real-valued time series data. To achieve this goal, the present paper proposes a different strategy to convert the exact prediction into the fuzzy domain. For this purpose, a non-parametric kernel-based statistical method was suggested and discussed. In this regard, flexible right and left spreads were introduced as some parametric forms and the fuzzy prediction scheme was constructed as LR-fuzzy numbers. A hybrid algorithm was also developed to evaluate the optimal values of the autoregressive order, bandwidth, and the unknown parameters of left and right spread. Four popular time series data sets were employed to examine the proposed algorithm. The accuracy of the proposed method was also evaluated in terms of some performance measures through its comparison with some common fuzzy time series models. The results indicated that the proposed fuzzy time series model is potentially effective in forecasting fuzzy time series data in real applications.