YIELD FORECASTING FOR FRUITS AND VEGETABLES USING RANDOM FOREST ON LARGE-SCALE AGROCLIMATIC DATA

Abstract

Accurate yield forecasting is essential for sustainable agriculture and climate adaptation, especially in regions exposed to extreme weather conditions such as Central Asia. This study investigates the use of machine learning models, such as Linear Regression (LR), Decision Tree (DT), Random Forest (RF), and Gradient Boosting (GB), for predicting fruit and vegetable yields based on large-scale agroclimatic data. After preprocessing and feature encoding, all models achieved high accuracy (R² > 0.90), with LR and GB slightly outperforming RF by less than 1%. Despite this small difference, RF demonstrated greater robustness and interpretability, remaining stable under data variability and missing values. Feature-importance analysis identified rainfall as the dominant predictor, followed by fertilizer use and irrigation. Temperature exhibited a nonlinear effect on yield, with productivity plateauing above 32–34 °C, highlighting the risk of heat stress in arid climates. These findings confirm that ensemble-based methods, particularly Random Forest, provide a practical balance between accuracy, stability, and explainability for climate-smart yield forecasting in real agricultural environments.