Heavy metals removal from wastewater using nanoporous adsorbent: Separation analysis via machine learning model

Abstract

This paper presents a comprehensive analysis of three predictive models, namely Multi-Layer Perceptron (MLP), LASSO, and Extreme Gradient Boosting (XGB) for estimating concentration (C) of a substance in a given dataset. Adsorption separation was considered for water treatment, and the models were employed for tracking the concentration variations of solute in the process. With x(m) and y(m) as input variables which are the location, and concentration (C) measured in mol/m³ as output, the dataset comprises more than 19,000 data points. The Fireworks Algorithm (FWA) was employed to perform hyper-parameter optimization for the models. Different metrics were utilized to gauge the proficiency of each model, such as the R2 (coefficient of determination) for both the training and testing datasets, RMSE, and MAE. Results indicate that the MLP model obtained the highest score in terms of R2, with values of 0.99751 for the training data and 0.99756 for the testing data, suggesting excellent predictive accuracy. The MLP model also demonstrated the lowest error rates, with an RMSE of 1.3937 E+00 and an MAE of 8.81875E-01. The results revealed that the employed machine learning models are well capable of predicting adsorption process when combined with mass transfer modeling and great accuracy can be obtained.