Neural intuitionistic fuzzy system with justified granularity

Abstract

Fuzzy systems are intensively investigated and extended to construct forecasting models. In particular, intuitionistic fuzzy sets are used to capture higher levels of uncertainty occurring in the modeled data. Neural networks are also used to reflect nonlinearity relationships frequently observed in time series. This paper proposes a new hybrid system merging fuzzy system with neural networks and an advanced optimization technique, the principle of justified granularity. Using this technique, we construct an innovative time-series forecasting model. In the experimental part of the paper, we demonstrate the advantages arising from applying the proposed approach to metal price forecasting. Finally, we provide evidence that the proposed model is competitive with the current state-of-the-art models for the forecasting horizons of one and five days.