Yapay Zeka Hızlı Bakış & Çıkarımlar
AI InsightThe findings suggest that agricultural input price policy should be integrated with climate risk management and that the identified lag structure can contribute to early warning systems.
📌 3 Maddede Bu Makale
Abstract
The main objective of this study is to determine the dynamics of agricultural input prices in Turkey. The study incorporates meteorological and macroeconomic data as dependent and explanatory variables. The data used in the research covers the period from February 2020 to April 2026. The ten subcategories of the Turkish Statistical Institute (TÜİK) Agricultural Input Price Index are used as dependent variables, while average temperature, total precipitation, temperature anomaly, USD/TRY exchange rate, and Brent crude oil price are used as explanatory variables. A separate XGBoost model was established for each category. Hyperparameter optimization with Optuna, time series cross-validation, and SHAP interpretability analysis were applied in the study. According to the findings, XGBoost outperforms Ordinary Least Squares (OLS) regression in all categories. The highest predictive power was obtained in the Other (R² = 0.532) and Energy (R² = 0.500) categories. Temperature anomaly and lagged precipitation variables provided significant SHAP contributions in many categories, especially energy, fertilizer, and animal feed. These results indicate that climate variables exhibit category-specific and lagged predictive associations with agricultural input price changes. On the other hand, the Seeds, Pesticides, Veterinary, Machinery Maintenance, and Building Maintenance categories produced negative R² values in all approaches. SHAP results for categories with negative out-of-sample R² values are interpreted cautiously as exploratory model-specific patterns, considering that the corresponding models do not exhibit reliable predictive performance. The findings suggest that agricultural input price policy should be integrated with climate risk management and that the identified lag structure can contribute to early warning systems.