Evolving ensemble of fuzzy models for multivariate time series prediction

Weather modeling and prediction has been quite a challenge over the years. Predictions based on climatic models whose dynamical behavior is nonlinear, nonstationary, and based on high order difference equations is a tough task and usually requires a demanding and non-intuitive tuning expertise. This paper suggests an ensemble of evolving fuzzy models for multivariate time series prediction. The proposed ensemble approach is able to model the weather dynamics from data streams concerning variables such as wet bulb temperature, atmospheric pressure, maximum temperature, and relative humidity of the air. The purpose is to predict rainfalls 5 days ahead while providing a linguistic description of the reasoning used to give the predictions. Empirical results show that the ensemble-based fuzzy evolving modeling approach outperforms other evolving approaches in terms of accurate predictions.