Machine learning strategy on activation energy of environmental heterogeneous reactions and its application to atmospheric formation of typical montmorillonite-bound phenoxy radicals

Heterogeneous transformation of organic pollutants into more toxic chemicals poses substantial health risks to humans. Activation energy is an important indicator that help us to understand transformation efficacy of environmental interfacial reactions. However, the determination of activation energ...

Full description

Saved in:
Bibliographic Details
Published in:The Science of the total environment 2023-10, Vol.895, p.165117-165117, Article 165117
Main Authors: Pan, Wenxiao, Chang, Jiamin, He, Shuming, Liu, Xian, Fu, Jianjie, Zhang, Aiqian
Format: Article
Language:English
Subjects:
Activation energy prediction
Heterogeneous reaction
Long-lived free radical formation
Machine learning framework
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Heterogeneous transformation of organic pollutants into more toxic chemicals poses substantial health risks to humans. Activation energy is an important indicator that help us to understand transformation efficacy of environmental interfacial reactions. However, the determination of activation energies for large numbers of pollutants using either the experimental or high-accuracy theoretical methods is expensive and time-consuming. Alternatively, the machine learning (ML) method shows the strength in predictive performance. In this study, using the formation of a typical montmorillonite-bound phenoxy radical as an example, a generalized ML framework RAPID was proposed for activation energy prediction of environmental interfacial reactions. Accordingly, an explainable ML model was developed to predict the activation energy via easily accessible properties of the cations and organics. The model developed by decision tree (DT) performed best with the lowest root-mean-squared error (RMSE = 0.22) and the highest coefficient of determination values (R2 score = 0.93), the underlying logic of which was well understood by combining model visualization and SHapley Additive exPlanations (SHAP) analysis. The performance and interpretability of the established model suggest that activation energies can be predicted by the well-designed ML strategy, and this would allow us to predict more heterogeneous transformation reactions in the environmental field. [Display omitted] •A dataset of activation energies for phenolic on MMT surfaces were obtained via DFT.•Accessible properties of phenolic and surface were used as inputs for ML models.•Decision tree model outperformed with lowest RMSE (0.22) and highest R2 (0.93).•Underlying logic understood via model visualization and SHAP analysis.•We establish a robust and explainable ML model to predict activation energy.
ISSN:0048-9697
1879-1026
DOI:10.1016/j.scitotenv.2023.165117