Advanced Deep Learning and Hybrid Models for Forecasting Rice Production in Indonesia

Forecasting rice production supplies in Indonesia can help state planning by predicting rice production so that the state is able to determine decisions in various aspects such as inventory levels, procurement planning, distribution scheduling, resource allocation, and price strategies that will be taken based on forecasting results using artificial intelligence. This research focuses on developing proposed machine learning, deep learning, and hybrid models so that researchers can determine their respective abilities in executing data they collect from official Indonesian sites. The methods used in the research include the ARIMA model, LSTM, and the ARIMA-LSTM hybrid model. Several evaluation methods used in research are R2, MAE, RMSE, MAPE, and SMAPE. Based on research that has been conducted, the LSTM-ARIMA hybrid residual model outperforms every evaluation result with R2: 0.97, MAE: 0.15, RMSE: 0.18, MAPE: 7.26%, SMAPE: 7.06%.