Developing machine learning algorithms to predict the dissolution of zinc oxide nanoparticles in aqueous environment

Abstract

Engineered nanoparticles (ENPs) are of particular concern due to their ubiquitous occurrence and potential to cause adverse effects on aquatic biota. Consequently, a comprehensive understanding of ENP interactions and the mechanisms that underpin their fate and behaviour in the aquatic system is important to support their long-term applications and protection of ecology. However, due to a wide range of physicochemical parameters, as well as possible dynamic interactions with natural colloid particles, it is not practical to undertake experimental testing for each variation of ENPs using different aquatic permutations. This study describes machine learning (ML) algorithms for prediction of nZnO dissolution in aquatic systems using experimental data. The input parameters with the highest correlation were size and pH. On the contrary, categorical input variables such as coating, coating type, salt, and NOM type had a low correlation. The random forest regression and the extreme gradient boost algorithms performed remarkably well, with coefficients of determination (R²) of 0.85 and 0.92, respectively. The least effective method was multiple linear regression, which had a root mean square error of 0.15 and an R² of 0.31. ML offers a convenient and low-cost approach for screening nZnO dissolution in aquatic systems.