Prediction of maize grain yield under phosphorus deficiency stress conditions with machine learning models

Maize is an important global crop, and phosphorus (P) affects its metabolic processes and performance. Accurate yield prediction is highly essential for agricultural planning and management. The present study investigated the application of machine learning models, such as support vector machines, multilayer perceptron, radial basis function, a hybrid model (i.e., grey wolf optimization-support vector regression [SVR-GWO]), and SVR to predict maize grain yield under normal and P deficiency stress conditions. A data set, including maize growth parameters and seed yield from controlled experiments of soil P levels, was used for this purpose. The inputs of each model were selected through variables entered into the multiple regression model. The performance of the models was evaluated using statistical criteria such as mean absolute percentage error, root mean square error, mean absolute error, and coefficient of determination. The SVR-GWO hybrid model outperformed other models and demonstrated its strength and accuracy in predicting maize grain yield under normal and P deficiency stress conditions. Sensitivity analysis identified the harvest index (HI) as the most important variable affecting seed yield. In addition, combining 100-grain weight, total plant weight (BY1), and HI as common inputs in the SVR-GWO model led to high accuracy and flexibility in predicting maize grain yield, reducing the need to measure more parameters for yield modeling. These results demonstrate that the SVR-GWO model serves as a robust tool for accurately estimating maize grain yield from a reduced set of harvest-stage traits. This highlights its practical utility for post-harvest yield-component analysis, rather than for prospective, early-season yield forecasting.