Forecasting Hazelnut Market in Türkiye Using DeepDenT: A Deep Learning-Based Approach

Abstract

This study analyzes long-term trends and future projections in Türkiye’s hazelnut sector using annual data on production, harvested area, and yield from 1961 to 2023. Forecasts for the period 2024–2028 were generated using the DeepDenT artificial neural network model, aiming to provide evidence-based input for agricultural planning and policy development. The model was trained on time series structures with statistically optimized lag selections, adopting a tailored approach for the temporal dynamics of each variable. The findings indicate a projected increase in both production volume and harvested area, while yield per hectare is expected to decline. This suggests a shift toward more efficient input use and technologically adapted farming practices. The model’s performance was evaluated using error metrics such as RMSE and MAPE, with the most accurate forecasts observed for harvested area. Yield predictions showed moderate accuracy, whereas production forecasts exhibited greater variability, indicating a need for model refinement in that aspect. A linear Granger causality test was conducted to examine potential dependencies among the variables, revealing no statistically significant linear relationships. This supports the appropriateness of forecasting each indicator independently. The findings aim to support producers, researchers, and policymakers in enhancing the resilience and sustainability of hazelnut production systems in Türkiye.