Machine learning-based prediction of potato yield using key agroclimatic and nutritional variables

Accurately predicting potato yield under tropical conditions remains a complex challenge due to the crop’s sensitivity to environmental and nutritional factors. This study evaluates the performance of five machine learning (ML) algorithms—Multiple Linear Regression (MLR), K-Nearest Neighbors (KNN), Support Vector Regression (SVR), eXtreme Gradient Boosting (XGB), and Artificial Neural Networks (ANN)—to predict potato yield using climate and foliar nutrient data collected from 21 field experiments across São Paulo State, Brazil. Climate variables were stratified by phenological stage, and variable selection was performed using the Random Forest algorithm to reduce dimensionality and enhance interpretability. Among the tested models, SVR achieved the best predictive performance (adjusted R² = 0.76), particularly when using the top 10 variables, which included precipitation, temperature, and concentrations of P, Ca, and N. These results highlight the potential of ML approaches to capture biologically meaningful interactions and improve yield forecasts, thereby contributing to more efficient decision-making in precision agriculture.