Assessment of machine learning-based methods predictive suitability for migration pollutants from microplastics degradation

The aim of the work was to assess the usefulness of machine learning in predicting the migration of pollutants from microplastics. The search for methods to reduce unnecessary laboratory analyzes is a necessary action both to protect the environment and from an economic perspective. Multiple regression, artificial neural networks, support vector method and random forest regression were used in the study to predict leaching of plasticizers and other contaminants from microplastics. The development of the methods were based on the results of laboratory tests obtained by the GC-MS method. The results obtained confirm the potential of artificial neural networks and the support vector method for effective modelling and prediction of chemical compounds leached from microplastics. Correlation results were obtained for the analyzed parameters between the data obtained in the model and laboratory data in the range of 0.96–0.98 and 0.93–0.99 for artificial neural networks and the support vector method, respectively. Multiple regression showed the lowest performance in all cases in predicting plastic phthalic acid esters (coefficient of determination (R2) in the range of 0.03–0.24). The results presented in this paper will provide new insight into the influence of different parameters and factors on the leaching of plastic additives. This information is necessary to assess the harmfulness of these materials. The collected data is unique on a global scale. For the first time, machine learning were used to predict the leaching rate of plasticizers from different polymers under different environmental conditions. The use of machine learning allows to reduce unnecessary laboratory tests and reduce costs and protect the environment. Currently, there are no research results in this field in the scientific literature.