Integrating experimental and python-based machine learning analyses to evaluate growth, morpho-physiological responses and nutrient dynamics of buxus species in response to sheep wool fertilization

Abstract Recycling agricultural by-products as organic fertilizers is an important strategy for sustainable plant production. This study investigated the effects of sheep wool waste as an organic fertilizer on plant growth, root architectural traits, nutrient dynamics, and chlorophyll content of B. balearica and B. sempervirens . Wool fertilizer was applied at four doses (0, 50, 100, and 200 g pot⁻¹), and growth, root architecture, chlorophyll content, and leaf and root nutrient concentrations were evaluated. The results revealed species-specific responses. In B. sempervirens , the 100 g treatment produced the most positive growth response, increasing shoot number by 13.4% compared to the control. In B. balearica , plant height was not significantly affected, but leaf development improved, with leaf length increasing by approximately 21% under the 200 g treatment. Wool application influenced nutrient accumulation in plant tissues. Leaf potassium concentrations increased by 48.6% in B. sempervirens and 128.2% in B. balearica , while chlorophyll content increased by 29.8% in B. balearica at 100 g. To identify key traits associated with plant responses, machine learning algorithms were applied using Python-based models. Five supervised machine learning algorithms were evaluated, and Random Forest was selected as the best-performing model for feature importance analysis and model interpretation. Morphological characteristics were the principal contributors to fertilizer treatment discrimination, followed by root architectural traits. These findings suggest that sheep wool fertilizer has potential as a sustainable organic amendment for ornamental boxwood production under greenhouse conditions. Integrating conventional statistical analyses with explainable machine learning provided complementary insights into plant responses and trait contributions.