We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
Machine Learning Approaches for Predicting Microplastic Removal
Summary
Researchers applied Bayesian optimization combined with machine learning models — including boosted regression trees, neural networks, and support vector regression — to predict microplastic removal during coagulation, finding the BOA-BRT hybrid outperformed conventional methods by up to 71%, with microplastic particle size identified as the most influential variable.
Microplastics (MPs), the newest type of pollution, are present almost everywhere in the world. This study investigated the possibility of using hybrid Bayesian optimization algorithm (BOA) and machine learning (ML) techniques (e.g., artificial neural network (ANN), boosted regression tree (BRT), and support vector regression (SVR)) to forecast the removal of MPs during the coagulation process for the first time. The independent variables, including polypropylene microplastic (PPMPs) size, pH, polyacrylamide (PAM), and polyaluminium chloride (PAC), were considered, while the MPs removal rate was the response variable. The results demonstrate the hybrid BOA-BRT model's superiority, with a high coefficient of determination (R2) and low mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error values, indicating its effectiveness in predicting MPs removal efficiency. The hybrid BOA-ANN model shows higher MAE, RMSE, and MAPE values than BOA-BRT. The statistical multiple linear regression (MLR) and hybrid BOA-SVR models also yielded comparable results, with R2 values of approximately 0.82 and 0.83, respectively. The performance of the best predictive BOA-BRT model was compared with the existing response surface methodology (RSM) model (Adib et al. in J Environ Health Sci Eng 20:565–577, 2022. https://doi.org/10.1007/s40201-022-00803-4 ). Regarding MAE, RMSE, and MAPE values, BOA-BRT outperformed the RSM with a performance enhancement of about 68%, 71%, and 63%, respectively. The model's generalization ability was tested with extra simulated data. Sensitivity analysis showed the relative importance of the input variables on MP removal rate in decreasing order as PPMPs size > pH > PAM dose > PAC dose. This study creates novel avenues for investigating different microplastic removal technologies.