Coupling Soil Testing Data with Machine Learning Algorithms to Predict Plant Responses to Phosphorus in Sub-Saharan Africa

Abstract Optimizing phosphorus (P) management in agriculture is critical for food security and sustainable development. In this study, using maize as a model crop, we propose and validate a machine learning–based framework that integrates soil testing data, management practices, and climatic variables to improve estimates of plant responses to P fertilization. We collected and organized data from the literature, trained and validated machine learning models (random forest, support vector regression, and k-nearest neighbors) integrating soil testing data with climate and management variables to predict the effect of P fertilization on plant relative yield under variable environmental conditions in Sub-Saharan Africa (SSA). Overall, the selected models were suitable for predicting plant responses to P across different regions of SSA. Random forest presented a better predictive capacity for relative maize yield in response to P fertilization. Integrating soil testing data with climate and management variables through machine learning permits predictive modeling of yield responses to phosphorus fertilization. This approach can also be adapted and expanded for other nutrients, crops, and regions.