Optimizing Rice Production Forecasts with Combined Time Series Models

Rice is a staple food for billions of people, making accurate forecasting of rice production crucial for decision-making, particularly for supporting the new capital city, Nusantara, in East Kalimantan. This study evaluates forecasting models for rice production using various univariate time series methods and their ensemble counterparts. Ensemble members were generated using univariate methods such as the Naïve Method, Simple Moving Average, Trend Analysis, Exponential Smoothing (ETS), and Autoregressive Integrated Moving Average (ARIMA). These were then combined using three techniques: averaging, Bates-Granger weighting, and Artificial Neural Networks (ANN). The findings reveal that the ANN-based ensemble model achieved the highest forecasting accuracy, outperforming both individual models and other ensemble methods. Additionally, the Bates-Granger weighted ensemble showed better performance than the simple averaging method, highlighting the benefit of performance-based weighting. Overall, ensemble models demonstrated superior accuracy over individual univariate models. Future research should explore incorporating more advanced machine learning techniques for ensemble generation and use encompassing tests to optimize forecast combinations.