Input Variables Selection Using Mutual Information for Neuro Fuzzy Modeling with the Application to Time Series Forecasting

Abstract

This paper presents a methodology to select input variables for time series prediction. A main motivation is to find some proper input variables which describe the time series dynamics properly. It is shown that even when the choice of input variables is confined to the lagged values of the process to be predicted, a nonlinear analysis of the most significant factors is crucial for improving the prediction quality. The proposed method is used to select the appropriate input variables for neuro fuzzy models utilized for time series prediction benchmark in NN3 competition as well as a second benchmark to show the generality of the claims. Results depict the effectiveness of the proposed method in proper input selection for neuro fuzzy models for prediction task.