Application of hybrid ARIMA and support vector regression model for improvement of time series forecasting

Accurate investigation related to the structure of time series plays an important role in increasing the accuracy of ARIMA forecasting. The aim of this research is to investigate the effect of modeling decomposition of linear and non linear parts of time series on ARIMA model results. The decomposition of wheat and maize yield time series (in Kermanshah and Esfahan provinces) in the linear part was related to ARIMA and in the non linear part was conducted with support vector regression (hybrid model). The kind of configuration of non linear part of hybrid model is more important for example in the maize time series of Kermanshah, the values of RMSE for configuration with residual was 1.52 and for time series configuration was 15.03. The decreasing of RMSE, MAE and UII for wheat time series of Esfahan with hybrid model was 45.94%, 52.29% and 46%, respectively which is indicative of hybrid model improvement. The value of GMER in all four time series was greater than one which indicates the overestimation of hybrid model. Comparison the average of each criteria with two models and crops in each province indicated the effect of climate on modeling process because the average of criteria in Esfahan province decreased rather to Kermanshah (RMSE decreasing= 24.72%, UII decreasing=12.24%). Therefore, decomposition of time series to linear and non linear parts of time series can increase the accuracy of ARIMA model results.