A COMPARISON OF TIME SERIES MODELS AND ARTIFICIAL NEURAL NETWORK MODELS FOR FORECASTING TÜRKİYE'S MONTHLY POULTRY MEAT EXPORTS TO IRAQ

Abstract

Accurate forecasting of agricultural exports is crucial for effective trade planning, particularly in markets exposed to volatility and political uncertainty. This study compares the forecasting performance of traditional time series models and artificial neural network (ANN) in predicting Türkiye’s monthly poultry meat exports to Iraq. Monthly export data from January 2010 to December 2020, obtained from UN Comtrade and the Turkish Statistical Institute (TURKSTAT), were analyzed using the Autoregressive Integrated Moving Average (ARIMA) model and a Feed-Forward Neural Network (FFNN) model. The ARIMA model was constructed following the Box–Jenkins methodology, while the ANN model was applied using supervised learning with lagged export values as inputs. Forecasting accuracy was evaluated using MAE, RMSE, AIC, and R². The results reveal that poultry meat exports exhibit nonlinear and volatile patterns, particularly during periods of regional instability. Empirical findings indicate that the FFNN model outperforms the ARIMA model, providing lower forecast errors and a better goodness of fit. These results suggest that ANN-based models offer a more reliable tool for forecasting poultry meat exports between Türkiye and Iraq and can support informed decision-making for exporters and policymakers.