Prediction of bisphenol A contamination in Canadian municipal wastewater

Bisphenol A (BPA) is one of the most common contaminants of emerging concerns (CECs), which pose a threat to human health. Conventional wastewater treatment plants (WWTPs) are considered as the major pathway of BPA entering the aqueous environment. To control and mitigate BPA contamination in the aquatic environment, predicting BPAs fate at WWTPs is critical. In this study, three machine learning models, including shared layer multi-task neural network (MLT-NN), genetic programming (GP), and extra trees (ET) are used to predict the effluent BPA concentration at twelve municipal WWTPs across Canada. Additionally, the theory of networks is adopted to analyze the interdependencies among the influencing factors of BPA removal. It is found that the proposed models can provide reasonable BPA effluent concentration predictions. They have advantages in alleviating data sparsity and imbalance, improving model interpretability, and measuring predictor importance, which is valuable for the modeling of BPA and many other CECs. The network analysis results imply there are moderate interdependencies among various influencing factors of BPA removal. Factors that significantly affect BPA effluent concentration and are thus important for BPA removal are identified. The results also show that BPA is unlikely to be removed at primary treatment plants, while BPA removal could be achieved through secondary or tertiary treatment. This study presents an integrated framework for the modeling and analysis of BPA at WWTPs, which can provide direct and robust decision support for the management of BPA as well as other emerging contaminants in municipal wastewater.