Efficient Prediction of Microplastic Counts from Mass Measurements

Abstract

Microplastics must be characterized and quantified to assess their impact. Current quantification procedures are time-consuming and rely on expensive equipment. This study evaluates the use of machine learning to estimate the number of microplastic particles based on aggregate particle weight measurements. Synthetic datasets are used to test the performance of linear regression, kernel ridge regression and decision trees. Kernel ridge regression achieves the strongest performance, and it is also tested with experimental datasets. The numerical results show that the algorithm is better at predicting the counts of larger and more homogenous samples, and that contamination by organics does not significantly increase error. In mixed samples, prediction error is lower for heavier particles, with an error rate comparable to or better than that of manual counting. Overall, the proposed method is faster, cheaper and easier than current approaches.SynopsisUsing generated and real datasets, this study demonstrates that Kernel Ridge Regression can estimate microplastic counts from weight measurements as accurately as traditional visual sorting techniques.