Application of multilayer perceptron network and random forest models for modelling the adsorption of chlorobenzene on a modified bentonite by intercalation with hexadecyltrimethyl ammonium (HDTMA)

Abstract

Prediction of adsorption capacity, one of the most important properties of any adsorbent-adsorbate system, is crucial for adsorption studies. In this investigation, two approaches such as multilayer perceptron (MLP) and random forest (RF) were used to predict the adsorption capacity of hexadecyltrimethyl ammonium modified bentonite to remove chlorobenzene (CB) from aqueous solution. The adsorption study was conducted in batch mode at different adsorption parameters. The results show that the adsorption processes were best described by Freundlich isotherm model, while the adsorption mechanism followed the pseudo-second order kinetics. It was observed that the structure of MLP model that give the best prediction consisted of three layers: input layer with four neurons, output layer with one neuron, and four neurons at hidden layer. The three important parameters for RF model were n tree = 500, m try = 1, and node size = 1. According to the results obtained, the MLP model provided slightly higher levels of accuracy with a consistently high coefficient of determination (R 2 = 0.996) and low root mean square error (RMSE = 0.00101) compared to RF model. (R 2 = 0.969, RMSE = 0.03008). Therefore, the initial concentration of CB with 35.29%, appeared to be the most influential parameter in the adsorption of CB on the modified bentonite.

A predictive analytics solution matching job seekers' talent and employers' demands based on machine learning

“…Artificial neural network models have been developed since 1980. The multilayer perceptron model is one of the most influential models (Bougdah et al, 2021).…”