Abstract
Meteorological conditions play an important role in determining soybean grain composition, although their relationships with chemical constituents are often complex and nonlinear. This study investigated the association between weather conditions and the proximate composition of soybean grains collected from commercial fields in Rio Grande do Sul during the 2024/2025 growing season. Protein, ash, lipids, fiber, and carbohydrates were quantified by near-infrared spectroscopy, while meteorological data covering the growing season were obtained for each sampling location. Linear regression analyses revealed weak linear associations, with coefficients of determination no greater than 0.202 for all evaluated components. To better represent these complex interactions, an artificial neural network classification framework was developed. Chemical composition data were partitioned into high and low groups using K-means clustering, and the resulting labels were used to train independent artificial neural network classifiers based on mean air temperature and weekly precipitation. After 300 independent executions, the proposed models achieved mean training accuracies above 99.95% and mean testing accuracies above 88.13% for all evaluated constituents. The best-performing models reached 100% accuracy on both the training and testing datasets, while analysis of variance confirmed statistically significant differences between the predicted groups for protein, lipids, fiber, and carbohydrates. The proposed methodology demonstrates that machine learning techniques can identify meteorological patterns associated with distinct soybean grain composition profiles, providing a complementary approach for evaluating environmental influences on grain quality.