A proposed technique for the problem of selecting the best forecasting model in time series : a case study

Abstract EN

Forecasting is considered as one of the essential goals regarding time series analysis, and forecasting accuracy decreases the risk possibility regarding decision making.

Whereat the best model to represent the time series data, might not be the same model used for forecasting.

The forecasting evaluation criterion used such as RMSE, MAPE, and MAE provides almost different results concerning one time series, which confuse the researcher to select the best model for forecasting.

Therefore this research deals with the problems of criterion differences results, that affects the evaluation performance to select the best model, and providing a statistical manner that employs the forecasting criterion results that are mentioned as a weighted mean for each model of ARIMA which is considered as a candidate model with the least weighted mean that provides the best forecasting performance.

This has been applied on the monthly time series for the water of the Tigris River (M-cu-m) that enters Mosul City for the period 1963-1995.

Meanwhile Box-Jenkins model shows SARIMA (1,1,2) * (3,1,1) 12Very encouraging forecasting results depending on the suggested manner compared with the rest of models, meanwhile the best model for representing the data is SARIMA (1,1,2) * (0,1,1) 12.