Predicting microplastic impacts on microalgae: A machine learning approach to understand dynamic interactions in aquatic ecosystems

Aquatic ecosystems are increasingly threatened by microplastic (MP) pollution, with microalgae, which are critical primary producers, exhibiting high sensitivity to MP-induced physiological stress. However, current models are inadequate for capturing the time-dependent dynamics of MP-microalgae interactions, particularly those shaped by environmental variability and the evolving physicochemical properties of MPs. In this study, we propose a comprehensive machine learning (ML) framework to model and predict the effects of MPs on microalgal growth. This study introduces a comprehensive machine learning (ML) framework to model and predict the impacts of MPs on microalgae growth. By integrating key MP characteristics (e.g., particle size, zeta potential, polymer type) with environmental parameters (e.g., temperature, light intensity), we constructed predictive models using six ML algorithms, including random forest and XGBoost, to evaluate both their predictive accuracy and explanatory power. Feature importance was quantified using SHAP analysis, which highlighted temperature, MP type, and zeta potential as dominant drivers of microalgal response. Furthermore, partial dependence plots (PDP) and individual conditional expectation (ICE) analyses revealed a temporal shift in influencing factors: environmental variables governed early-stage growth, whereas MP-specific properties became increasingly impactful over time. Among all models, XGBoost consistently demonstrated superior performance across time intervals. These results offer novel insights into the complex, time-resolved interplay between MPs, environmental stressors, and microalgae, informing strategies for managing microplastic pollution in aquatic environments.