<![CDATA[An accurate forecasting model for a time series data is still difficult to obtain if the data is complex. This study aims to analyze and make the model of import export data forecasting with the combined method ARIMA - Neural Networks. This method is expected to improve NN's ability to complex problems and improve forecasting accuracy. The forecasting model obtained is used to predict the value of import-export in the next period. From the available data, ARIMA forecasting model for export value is ARIMA (1,1,12) with error 0,968 and forecasting model of NN with sigmoid bipolar gives error 0,180732 while NN model with semilinier gives error 0,081521 . For import value, obtained ARIMA (0, 1, 0) model with error 0,971 and forecasting model of NN with sigmoid bipolar gives error 1,437723 while model of NN with semilinier gives error 0,957831. Based on these results, a combined forecasting model of ARIMA and Neural Network with a semilinier activation function will be performed because it has a smaller error value compared to the sigmoid bipolar activation function. The ARIMA- NNforecasting model with the semilinier activation function yield error 0.046010 for the export value data and 1.081964 for the import value data.]]>
Copyrights © 2017
