<![CDATA[This study compares the performance of the Autoregressive Integrated Moving Average (ARIMA) model and an Artificial Neural Network (ANN) in forecasting annual palm oil production in Kampar Regency, using a univariate time series covering the period from 2013 to 2024. The forecasting aim is to support regional agricultural planning and decision-making in one of Riau Province’s key oil palm-producing regions. The ARIMA model was developed using the Box–Jenkins approach, which involves stationarity testing, optimal model identification, parameter estimation, and residual diagnostics, including ACF/PACF, Shapiro–Wilk, Jarque–Bera, and Ljung–Box tests. A feedforward ANN with three lagged inputs, five hidden neurons, sigmoid activation, and backpropagation training was constructed for comparison. Model performance was evaluated using RMSE, MAPE, and R². The results indicate that the ARIMA (1,1,1) model yields more stable and reliable forecasts, with diagnostic tests confirming white noise residuals and no significant autocorrelation. Conversely, the ANN model produced higher errors and indications of overfitting, likely due to the limited number of observations and the sharp increase in production recorded in the final data year. While ANN captured a stronger upward trend, which may represent an optimistic scenario, ARIMA provided more conservative and statistically valid forecasts under constrained data conditions. Overall, the ARIMA(1,1,1) model proved more suitable for the short univariate palm oil production series, yielding lower forecasting errors (RMSE = 273.88; MAPE = 8.92%) than the ANN model (RMSE = 283.53; MAPE = 9.03%).]]>
Copyrights © 2025
