FORECASTING INDONESIA COMPOSITE INDEX USING HYBRID AUTOREGRESSIVE INTEGRATED MOVING AVERAGE-DOUBLE RANDOM FOREST MODEL

Abstract

Modeling time series data using autoregressive integrated moving average (ARIMA) has been widely discussed. However, this has limitations in that it can only handle linear data. Machine learning is one of the alternative approaches that can solve this limitation since this method can handle nonlinear cases. Double random forest (DRF) is considered a supervised learning method that can solve regression problems. This research provides a novel hybrid forecasting framework combining ARIMA and DRF, designed to model both linear and nonlinear behaviors, and provide more accurate predictions for volatile financial data like the Indonesia Composite Index (ICI). Previous studies have not examined the performance of the hybrid ARIMA-DRF model. In this study, the performance of ARIMA, DRF, and the hybrid ARIMA-DRF models is compared using ICI data obtained from Bank Indonesia’s website. ICI has nonstationary and nonlinear characteristics. This made the ICI data suitable to be modeled using the hybrid ARIMA-DRF model. The comparison results indicate that the hybrid ARIMA-DRF model outperforms the independent ARIMA and DRF models, with a value of its mean absolute percentage error is 4.17%. Therefore, forecasting the future value of ICI data was done by using a hybrid ARIMA-DRF model with forecasting periods from October 2023 to September 2024. The forecasting results show that ICI values fluctuate over the forecasting periods, hence the authority might use the pattern to predict the ICI behaviors and take further decisions. While the forecasting results offer valuable insights for decision-making, this study has limitations as it does not incorporate external macroeconomic variables that may influence ICI behavior.