Comparison of Linear Regression, ARIMA, Simple Exponential Smoothing, Hybrid ARIMA-LSTM, and EWMA in Forecasting Commodity Prices

In this study, we compare the performance of both hybrid and non-hybrid forecasting models, explicitly focusing on Linear Regression, ARIMA, Simple Exponential Smoothing, Hybrid ARIMA-LSTM, and EWMA in predicting commodity prices within the volatile market of Central Java, Indonesia. The primary objective is to evaluate which hybrid and non-hybrid models provide the most accurate and reliable forecasts under various conditions. Analyzing daily price data from the SiHaTi platform, an official service provided by Bank Indonesia, the Hybrid ARIMA-LSTM model emerges as the most accurate, achieving a forecast accuracy of 92.5%, compared to the 78.3% and 84.7% accuracies of Linear Regression and ARIMA, respectively. These findings underline the potential advantages of combining machine learning with statistical methods to improve predictions in dynamic market conditions, providing invaluable insights for policymakers and market analysts. However, it should be noted that only one hybrid model was compared, and future research should explore multiple hybrid models to ensure a comprehensive evaluation of their effectiveness.