A Modified Approach by Using Prediction to Build a Best Threshold in ARX Model with Practical Application

Abstract

The proposal of nonlinear models is one of the most important methods in time series analysis, which has a wide potential for predicting various phenomena, including physical, engineering and economic, by studying the characteristics of random disturbances in order to arrive at accurate predictions.

In this, the autoregressive model with exogenous variable was built using a threshold as the first method, using two proposed approaches that were used to determine the best cutting point of [the predictability forward (forecasting) and the predictability in the time series (prediction), through the threshold point indicator]. B-J seasonal models are used as a second method based on the principle of the two proposed approaches in determining the best seasonal model. Then they are compared with the obtained models from two methods that mentioned above of the two approaches within a group of the criteria as AIC, MDL, Loss Function, BIC, FPE, MSE, in addition the proposed weighted comparison criteria to determine the best model for representing the wind speed data as input variable, soil and dust as an output variable, in Baghdad Station from January 1956 to December 2012.