Exploring Drought Response: Machine-Learning-Based Classification of Rice Tolerance Using Root and Physiological Traits

Drought is a key limitation for rice productivity. While oxidative stress markers like hydrogen peroxide (H2O2) are important for drought adaptation, the predictive value of combining root anatomical and physiological traits is underexplored. We assessed 20 rice cultivars under drought and control conditions using a random forest, a multi-layer perceptron, and a SHAP-optimized stacking ensemble. The stacking ensemble achieved the highest classification accuracy (81.8%) and identified hydrogen peroxide, relative water content, and endodermis inner circumference as key predictors. SHAP analysis revealed important interactions between root anatomical and physiological traits, providing new biological insights into drought tolerance. Our integrative approach, supported by robust cross-validation, improves predictive power and transparency for breeding drought-resilient rice cultivars.