Variable Scaling for Time Series Prediction: Application to the ESTSP´07 and the NN3 Forecasting Competitions

In this paper, variable selection and variable scaling are used in order to select the best regressor for the problem of time series prediction. Direct prediction methodology is used instead of the classic recursive methodology. Least Squares Support Vector Machines (LS-SVM) and K-NN approximator are used in order to avoid local minimal in the training phase of the model. The global methodology is applied to the ESTSP´07 competition dataset and the dataset B of the NN3 Forecasting Competition.