Comparison of ARIMA and NNAR Model for Rice Production Forecasting

Abstract

Accurate forecasting of rice production is crucial for ensuring future food security in specific regions. This study aimed to compare the predictive capabilities of two distinct time series models, Autoregressive Integrated Moving Average (ARIMA) and Neural Network Autoregression (NNAR), in forecasting rice production. The selection of input variables for the NNAR model was aligned with the autoregressive components of the ARIMA model. Applying annual rice production data for the East Kalimantan Province from 1994 to 2023, this study evaluated forecasting accuracy through the Symmetric Mean Percentage Error (SMAPE). The results demonstrated the superior performance of the ARIMA ([2],0,0) model compared to the NNAR (2, 8) model. The advantage of the ARIMA ([2], 0, 0) model is attributed to the limited amount of data used, making simpler models like ARIMA more suitable for this case. Future research could develop more comprehensive forecasting models by incorporating additional factors such as weather data. Furthermore, the adopting of hybrid models combining ARIMA and NNAR, along with an increase in the quantity and diversity of data, could be potential steps towards enhancing the accuracy of rice production forecasting in the future.