Comparative analysis of prediction models for Turkey's sunflower oil imports

Abstract

This study examines the comparative performance of traditional statistical and machine learning (ML) techniques in forecasting Turkey's sunflower oil imports. The analysis includes Seasonal ARIMA (SARIMA), ARIMAX, Random Forest Regression (RFR), Support Vector Machines (SVM), and Multiple Linear Regression (MLR). The performance of the models is evaluated across short-, medium-, and long-term horizons using a dataset spanning 19 years (2004–2023) and performance metrics such as Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and Theil's U-statistic (THEIL). Results reveal that RFR consistently outperforms other models due to its ability to handle complex datasets and nonlinear relationships. While SARIMA excels in short-term predictions, ARIMAX effectively captures medium- and long-term trends. This study offers actionable insights for policymakers, highlighting the potential of machine learning (ML) techniques to enhance agricultural trade strategies and mitigate risks associated with import dependencies.