Forecasting Indonesia^s Export Value using The Hybrid Method of ARIMAX-FFNN and Extreme Gradient Boosting (XGBoost) Andra Citta Passadhi Arya, Heri Kuswanto, Kartika Fithriasari
Department of Statistics, Institut Teknologi Sepuluh Nopember (ITS), Sukolilo, Surabaya, 60111, Indonesia
Abstract
Accurate forecasting of Indonesia^s export value has an important urgency as a reference for formulating national economic growth targets. The value of Indonesia^s exports is believed to be influenced by various external factors, including the exchange rate of the rupiah and import values. Therefore, it is important to model the forecast of Indonesia^s export value by incorporating these factors. The forecasting method used is a hybrid approach that combines two models, a linear and a nonlinear model, with the aim of producing more accurate predictions. The first stage involves linear modeling using the Autoregressive Integrated Moving Average with Exogenous Variables (ARIMAX) model. Subsequently, in the second stage, hybrid modeling is conducted by utilizing the residual input from the linear model and employing a machine learning approach known as the Feed Forward Neural Network (FFNN) to capture nonlinear patterns. In addition to the hybrid modeling, another machine learning algorithm, Extreme Gradient Boosting (XGBoost), is also employed. The aim of this research is to compare the accuracy of the hybrid ARIMAX-FFNN and XGBoost forecasting models. By doing so, it aims to determine which model performs better in terms of accuracy. The selection of the best model is based on the smallest values of Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE).
