A Comparative Analysis of Hybrid Approach for Rice Production Forecasting

Accurate forecasting of rice production is essential for ensuring food security, stabilizing prices, and guiding agricultural policy decisions. This study develops a hybrid forecasting model by combining linear time series models (Linear Trend and Error, Trend, Seasonal (ETS)) with Neural Network Autoregression (NNAR). Using annual rice production data from East Kalimantan Province, Indonesia (1995–2023), the model's accuracy was evaluated using the Symmetric Mean Absolute Percentage Error (SMAPE). The results show that hybrid models consistently outperform their individual components, with the ETS-NNAR model achieving the highest accuracy. Among individual models, the ETS model demonstrated better performance than the Linear Trend model, suggesting a correlation between the accuracy of the individual model and its hybrid counterpart. These findings imply that improving the baseline model's accuracy could further enhance the hybrid model's performance. Future research should consider integrating alternative neural network architectures, such as deep learning and Long Short-Term Memory (LSTM) networks, and other machine learning techniques into hybrid frameworks to further improve rice production forecasting.