ABSTRACT
Methylene blue (MB) is an important compound in textile and wood processing industries as well as in medical research for combating malaria parasites. Despite these versatilities, direct contact with human beings results in adverse health challenges, and contamination of water bodies affects aquatic biotas. Hence, it is important to treat MB-contaminated wastewaters before disposal into water bodies. Adsorption, which depends on some parameters, proves to be an easy, cheap, and efficient technique to remove pollutants in wastewater. However, investigating these parameters experimentally is a laborious, expensive, and time-consuming process whose efficiency is limited by the conditions imposed on the experiments. Herein, we developed polynomial multiple linear regression (MLR) and the three other machine learning models to study the interplay of five adsorption parameters (descriptors) and their effects on the removal of methylene blue from water using aluminized activated carbon (Al-AC). The optimized machine learning models, that is random forest (R = 0.9905), support vector regression (R = 0.9946), and multilayer perceptron (R = 0.9993), outperformed the best MLR model (R = 0.9845) by small margins. High statistical R and low error values are not enough to satisfactorily classify a model. Hence, the generalizability of the models was further determined under different experimental conditions, and the order of predictive accuracy of the models was established as ANN > SVR > RF > 2-degree MLR. Aluminum loading, adsorbent dosage, and initial adsorbate concentration are the most important factors affecting MB removal. The removal efficiency, which could reach 99.9% at optimum conditions, does not depend on the temperature thus eliminating the need to install temperature control apparatus for practical setup.
