A New ANFIS Synthesis Approach for Time Series Forecasting

ANFIS networks are neural models particularly suited to the solution of time series forecasting problems, which can be considered as function approximation problems whose inputs are determined by using past samples of the sequence to be predicted. In this context, clustering procedures represent a straightforward approach to the synthesis of ANFIS networks. The use of a clustering procedure, working in the conjunct input-output space of data, is proposed in the paper. Simulation tests and comparisons with other prediction techniques are discussed for validating the proposed synthesis approach. In particular, we consider the prediction of environmental data sequences, which are often characterized by a chaotic behavior. Consequently, well-known embedding techniques are used for solving the forecasting problems by means of ANFIS networks.