Unraveling the Context-Dependent Effects of Microplastics on Soil Enzymatic Activities via Interpretable Machine Learning

Microplastics (MPs) increasingly accumulate in agricultural soils, yet their effects on soil enzymatic activity remain heterogeneous. Here, a literature-derived dataset was analyzed using an interpretable machine learning framework. eXtreme Gradient Boosting (XGBoost) and Random Forest (RF) captured non-linear patterns in enzymatic responses to MP characteristics. Particle size showed the strongest model-based contribution, with a dataset-derived size-sensitive range of 200–600 μm associated with more pronounced predicted inhibition. Dosage showed a plateau-like pattern above approximately 100 g/kg, while exposure time exhibited non-linear variation across the observed duration. SHapley Additive exPlanations (SHAP) analysis further characterized the relative contributions of individual features to model predictions. Overall, the results provide quantitative, data-driven insights into heterogeneous MP-soil enzyme associations and highlight particle size as an important factor for further experimental investigation and risk assessment.