Structural design of cellulose derivative-modified chitosan adsorbents for arsenic removal: Machine learning modeling, Box-Behnken design, Optuna hyperparameter tuning, and molecular dynamics

Abstract

Heavy metal contamination poses significant environmental and public health risks, necessitating the development of effective remediation strategies. Although cellulose derivatives are widely used to modify adsorbents, hydroxypropyl cellulose (HPC) has not yet been investigated for functionalizing adsorbents in heavy metal removal. This study aimed to enhance As(III) adsorption performance by optimizing the structural properties of the hydroxypropyl cellulose-modified chitosan/polyvinyl alcohol (HPC/CP) adsorbent using the Box-Behnken design (BBD) and several machine learning (ML) algorithms. The optimization process focused on four key structural parameters: chitosan (CS) content, HPC content, polyvinyl alcohol (PVA) content, and the thickness of the adsorbent. The random forest (RF) ML algorithm confirmed the highest predictive accuracy (R² = 0.99, MSE = 2.31) after hyperparameter tuning (max_depth = 38, min_samples_leaf = 29, min_samples_split = 13, n_estimators = 470). The optimal adsorbent composition was identified as 5 wt% CS, 4 wt% HPC, 12 wt% PVA, and 300 µm thickness. The optimized adsorbent achieved an experimental As(III) removal efficiency of 98.6 %, significantly higher than the unoptimized adsorbent (65.3 %). The RF-predicted removal (96.1 %) closely aligned with experimental results and outperformed other ML models. Molecular dynamics simulations further validated the structural stability of the As(III)/HPC/CP complex through reduced energy fluctuations. HPC functionalization, combined with ML-driven optimization, significantly enhances the As(III) removal efficiency of CS-based adsorbents.