Comparative Study of Statistical, Machine Learning, and Deep Learning for Rice Retail Price Forecasting in West Java

This research investigates the effectiveness of machine learning (ML), deep learning (DL), and statistical algorithms for forecasting rice retail prices in West Java, Indonesia. Employing data from 2021 to 2023, this study evaluates Random Forest, LightGBM, XGBoost, LSTM, and SARIMAX models. ML algorithms displayed high error rates with RMSE values exceeding 1100, indicating challenges with complex datasets. In contrast, LSTM models demonstrated robust performance with a substantially lower RMSE of 129.67, but LSTM model's low error rates was achieved by not involving other exogenous variables, focusing solely on historical price data. The SARIMAX model, while less effective in direct forecasting with an RMSE of 799.82, effectively utilized exogenous variables to encapsulate seasonal trends, offering valuable insights into supply chain dynamics affecting rice prices.