Predicting Heavy Metals Adsorption on Microplastics and Unraveling the Adsorption Mechanism with Machine Learning Methods

Abstract

Microplastics (MPs) are widely present in aquatic environment and easily interact with heavy metals (HMs) through adsorption, affecting their fate and ecological risks. However, elucidating the adsorption behaviors and mechanisms is challenging through laboratory methods owing to the difference in MPs properties and HMs species as well as complex environmental conditions. Herein, traditional regression, machine learning (ML), and deep learning models were employed and optimized to predict HMs adsorption by MPs. Random forest (RF) possessed superior prediction performance with the testing R2 and root mean square error (RMSE) of 0.93 and 0.04. Based on shapley additive explanations (SHAP) and partial dependence plots, the environmental conditions exhibited the greatest impact on adsorption (63.0%), followed by physiochemical characteristics of MPs (27.4%) and chemical properties of HMs (9.6%). The influence of solution pH, atomic mass of HMs, salinity, and aging degree of MPs suggests electrostatic interaction and complexation might dominate HMs adsorption. A web application was developed to allow the users to make rough predictions rapidly through a feasible approach. This work provides a novel ML-based perspective on understanding the interaction between HMs and MPs, which helps evaluate their ecological risks in aquatic environment.

Keywords: microplastics, heavy metals, machine learning, Adsorption, random forest