Ensemble hybrid machine learning to simulate dye/divalent salt fractionation using a loose nanofiltration membrane

Abstract

The escalating quantity of wastewater from multiple sources has raised concerns about both water reuse and environmental preservation. Therefore, there is a pressing need for intelligent tools that can aid in comprehending the intricate process of removing dyes and salts from wastewater beyond membrane technology. This study introduces novel standalone hybrid models that integrate an improved nonlinear ensemble approach to model the fractionation of dye and salt rejection (RJDS) (%) based on established experimental data. Using linear sensitivity analysis, two model combinations were identified based on different input variables: M1 (R = 52%, T = 50%, and P = 61%) and M2 (R = 52%, T = 50%, P = 61%, F = 71%, and RJ = 83%). These combinations were incorporated into hybrid neuro-fuzzy (NF) and least square support vector machine (LSSVM) models. The standalone and improved ensemble models were evaluated using several performance criteria, such as MSE, MAE, MAPE, RMSE, and PBAIS. The predictive outcomes demonstrated that NF-M2 outperformed all other models, with an MAE of 0.0002 and an RMSE of 0.0003. Similarly, the ensemble results indicated a significant improvement over the individual models. The study's findings demonstrate the reliability of intelligent tools for modelling RJDS (%) and serving as decision-making performance analysis tools. The proposed approach offers a novel, efficient and reliable technique for understanding and predicting dye and salt rejection in wastewater.