Pesticide contamination presents a considerable threat to human health and ecosystems, and photocatalysis is commonly utilized for pesticide breakdown. This research explores the forecasting abilities of machine learning models in predicting the photodegradation of pesticides with ZnO-based photocatalysts in water. By utilizing a comprehensive dataset derived from existing literature, the key physicochemical and process parameters, including light source type, dopant-to-Zn mass ratio, pesticide concentration, solution pH, and irradiation duration, were examined. Several machine learning techniques are employed, ranging from classic models of linear regression and decision trees to advanced artificial neural networks (ANN), CatBoost, and ensemble learning strategies. The performance of the models was assessed through standard evaluation criteria, namely the coefficient of determination (R2), mean squared error (MSE), and mean relative deviation (MRD), which together provide a comprehensive measure of predictive accuracy and reliability. The results reveal that ANN and CatBoost models outperform simpler models, achieving high R2 values (0.9234 and 0.9262, respectively) and low MSEs (40.67 and 39.16). Through advanced visual techniques, it is confirmed that ANN and CatBoost exhibit superior predictive accuracy and robustness, with minimal prediction errors. Additionally, the Shapley Additive exPlanations (SHAP) method is hired to understand feature significance, revealing that irradiation duration and initial pesticide concentration are the most influential factors in photodegradation. This work provides insights into optimizing pesticide photodegradation processes and emphasizes the utility of data-driven models in environmental remediation. In practice, these models can support the design of more efficient water treatment protocols, thereby contributing to improved public health and ecological safety.
扫码关注我们
求助内容:
应助结果提醒方式:
