Machine Learning-Based Evaluation of Nutrient Distribution in Grafted Cucumber Plants

This study investigated the effects of different grafting combinations on nutrient (Mg, S, Fe, B) accumulation and transport in cucumber plants, combining conventional analyses with machine learning. Greenhouse experiments compared non-grafted, self-grafted, and rootstock-grafted (TZ148, Cremna, Maximus) plants. Self-grafted plants accumulated the highest root Mg (0.095 µg/g), while non-grafted plants retained excessive Fe in roots (15.80 µg/g). In contrast, Maximus rootstock reduced root Fe (7.68 µg/g) and enhanced fruit Mg (0.414 µg/g), Fe (22.91 µg/g), and B (3.63 µg/g), while Cremna promoted higher S in fruit (0.640 µg/g). Random Forest models classified grafting combinations based on tissue nutrient profiles, identifying S, B, and Mg as the most discriminative features, supported by SHAP and PCA analyses. Overall, results show that grafting—especially with Maximus and Cremna rootstocks—optimizes nutrient allocation and fruit mineral quality. Integrating machine learning with physiological data provides a robust approach for improving rootstock-scion selection in sustainable cucumber production.