Growth, nutrient dynamics, and root architecture responses of Buxus sempervirens to chemical and biofertilizer applications: an integrated experimental and machine learning approach

This study evaluated the effects of different fertilizer applications (nitrogen, potassium, vermicompost, bacteria, mycorrhizal, and sheep wool) on plant development, root morphology, nutrient content, and chlorophyll content of Buxus sempervirens. The aim was to determine the effects of different fertilizer treatments on plant growth and nutrient dynamics and to evaluate the ability of machine learning classifiers to distinguish fertilizer treatments based on plant growth characteristics. Nitrogen was the most effective treatment, increasing plant height (25%), plant width (34%), shoot length (29%), and leaf length (18%) relative to the control. Shoot number was also numerically higher under nitrogen (51%), although the overall treatment effect was not significant. Potassium was the second most effective, improving plant height (25%), plant width (33%), and shoot length (9%) compared with the control. Nitrogen also had the strongest impact on root architecture, increasing root length (45%), root surface area (81%), and root volume (119%). The number of root forks was numerically higher under nitrogen (38%), although the overall treatment effect was not significant. Root nutrient analysis showed that treatment effects were significant for root N and K concentrations, whereas no significant treatment effects were detected for the other measured root nutrients. In leaves, treatment effects were significant for Mg, P, K, Fe, and Mn, whereas Ca, Zn, N, and Cu were not significantly affected overall. Chlorophyll content varied according to leaf age and measurement period, with nitrogen producing the highest mean value among young leaves in June, although the overall treatment effect was not statistically significant. Five machine learning classifiers were evaluated using six aboveground morphological characteristics. J48 showed the highest hold-out classification performance, whereas Naïve Bayes showed the highest five-fold cross-validation accuracy. Random Forest feature-importance and SHAP analyses indicated that plant size and shoot-development characteristics contributed more strongly to model output than leaf dimensions. Nitrogen was the most effective treatment for enhancing growth and root development, while sheep wool notably influenced selected nutrient responses. Fertilizer effects were trait-specific and varied among growth, root, chlorophyll, and nutrient characteristics. The machine learning analysis identified treatment-related morphological patterns but showed limited generalization to previously unseen observations. These findings indicate that combining conventional plant and nutrient analyses with interpretable machine learning approaches may provide complementary information for evaluating fertilizer responses in ornamental plants. • Nitrogen was the most effective treatment, significantly improving key plant growth and root traits in Buxus sempervirens. • Root traits showed strong associations with nutrient dynamics, indicating their key role in plant performance. • Sheep wool showed notable effects on selected nutrient responses, particularly potassium, iron, and manganese. • Nitrogen and potassium treatments produced the highest chlorophyll responses across measurement periods. • Machine learning models identified morphological traits contributing to fertilizer treatment classification.