Authors: Ambuj Kumar Misra
Abstract: Accurate prediction of wheat yield is essential for ensuring global food security and enabling data-driven agricultural policy decisions. This study presents a comparative analysis of two prom-inent ensemble machine learning algorithms—Random Forest (RF) and Extreme Gradient Boost-ing (XGBoost)—applied to wheat yield prediction using publicly available datasets from the United States Department of Agriculture (USDA), the National Oceanic and Atmospheric Ad-ministration (NOAA), and NASA remote sensing archives. A rich feature set spanning climatic variables, soil characteristics, spectral vegetation indices (NDVI), and growing degree days (GDD) was engineered from these sources. After rigorous preprocessing, 80% of the data was used for training and 20% for testing. XGBoost outperformed Random Forest, achieving an R² of 0.91, an RMSE of 3.47 bu/ac, and a MAE of 2.68 bu/ac, compared to 0.87, 4.21 bu/ac, and 3.15 bu/ac for RF. Both models substantially surpassed a linear regression baseline (R² = 0.74). Feature importance analysis consistently ranked maximum temperature, solar radiation, and NDVI as the most influential predictors. These findings underscore the potential of tree-based ensemble methods to deliver high-precision, scalable yield forecasts with freely accessible data, offering a cost-effective tool for agronomists, policymakers, and food supply chain planners.
DOI: http://doi.org/10.5281/zenodo.21452417
