Machine learning-based prediction and optimization of effective removal of tetracycline from wastewater using magnetic MoS2/polyglucosamine/β-cyclodextrin nanocomposite

Abstract

Background

The growing amount of tetracycline (TC) antibiotics in wastewater has given a warrant to major response due to the need of efficient removal strategies, which this research presents on the application of a magnetic MoS2 /polyglucosamine/ β_cyclodextrin (MPCN) composite nanocomposite as a quick and effective adsorption agent TC, and model the process via machine-learning sequence modelling.

Methods

It based the model formulation on two machine learning algorithms Lasso Regression (LR) and Polynomial Regression (PR) in predicting removal efficiency, using all essential key operational parameters to include pH (X₁), initial TC concentration (X₂), contact time (X₃), and adsorbent dosage (X₄). Particle Swarm Optimization (PSO) optimized work conditions (pH = 7, concentration = 1 mg L⁻¹, contact time = 90 min, dosage = 1 g L⁻¹) to improve model performance and accuracy.

Significant Findings

Maximum removal efficiency reached 94.7 %. The high performance of the multinomial regression model on the training data was clearly evident. Consequently, with regard to the training data, the model fitted very well and as a result R² and MAE were 0.91 and 2.85 % respectively. In addition, the high overfitting of the test data compared to the training data with an RMSE value of 6.23 % was a strong reason that the model performed much better on the training data. LR training accuracy was higher than PR in generalization (R² = 0.85, MAE = 3.42 %), and RMSE was lower by 18.3 % across the models. PSO had an improved accuracy (Average Error Reduction) when it came to modeling in practical application as compared to PR and LR.