Adaptive Multidimensional Neuro-Fuzzy Inference System for Time Series Prediction

This paper introduces a novel approach to forecast nonlinear time series using an adaptive multidimensional neuro-fuzzy inference system (AMNFIS), developed originally for processes control. In relation to other neuro-fuzzy systems, AMNFIS has a lower number of parameters avoiding the course of dimensionality problem. In addition, several strategies for fitting and model specification are discussed. In this paper, AMNFIS is used to forecast two well-known nonlinear time series and the results are compared against the forecasts obtained using the ARIMA approach and artificial neural networks. Empirical evidences indicate that AMNFIS is more accurate for forecasting the considered time series than the other two models.