# Activated carbon from banana peels for alizarin removal: understanding the adsorption process through isotherms, kinetics, and predictive modeling

## TL;DR

This paper explores using banana peel-based activated carbon to remove alizarin, combining adsorption studies and predictive models to optimize the process.

## Contribution

The study introduces a novel dual approach combining banana peel-derived activated carbon with mathematical and neural network models for alizarin removal.

## Key findings

- Pseudo-second-order kinetic model best describes chemisorption mechanism.
- Freundlich isotherm indicates multilayer adsorption on activated carbon surfaces.
- Artificial neural networks outperformed traditional models with 96.26% predictive accuracy.

## Abstract

A low-cost and efficient activated carbon was prepared from banana peels for the removal of alizarin. Non-linear kinetic studies, isotherm models, and predictive models were used to study the adsorption process. The choice of the kinetic studies and isotherm models was based on the error functions of R2, Adj.R2, chi-square, SSE and ARE. The kinetic studies using intraparticle diffusion, pseudo-first-order, Elovich, and pseudo-second-order models fit the data well. It was concluded that the pseudo-second-order model best fitted the data, thus the mechanism was by chemisorption. Isotherm studies indicated that the Langmuir, Temkin, Dubinin-Radushkevich, and Freundlich models all describe the adsorption process. The mode was best described by the Freundlich model, thus, adsorption occurred on multilayer surfaces. The quadratic model, with a high R2 value of 0.9740, accurately predicted the removal efficiency and identified dosage and concentration as the most significant factors. The optimized conditions were found to be 3.05 min, a pH of 5.55, a dosage of 0.014 g, and a concentration of 21.50 ppm, which resulted in a maximum removal efficiency of 92.18%. The artificial neural networks with a predictive capability (96.26%) and a correlation coefficient of 0.99999 for both training and validation sets was superior to the central composite design. This was confirmed through the comparison of the residual errors and the statistical error functions of SSE, MSE, RMSE and MAE. This study shows a dual approach of coupling activated carbon from banana peels with mathematical models to understand the adsorption process.