Artificial Intelligence-Assisted Prediction of Amoxicillin Removal from Wastewater Using Biomass-Derived Activated Carbons

Abstract

The presence of antibiotics in wastewater represents a pressing environmental and public health concern, highlighting the need for efficient and sustainable removal strategies. This study presents a comprehensive evaluation of amoxicillin (AMX) adsorption using three biomass-derived activated carbons prepared from walnut shells (WNSAC), pistachio shells (PSSAC), and pine nut shells (PNSAC). The adsorbents were characterized in terms of surface area, pore structure, surface chemistry, and crystallinity. Batch adsorption experiments were conducted under varying conditions of pH, initial AMX concentration, adsorbent dosage, contact time, and temperature. The adsorption kinetics followed the pseudo second order model, and the equilibrium data were best described by the Toth isotherm, indicating multilayer adsorption behaviour. Thermodynamic analysis confirmed that the process was spontaneous and exothermic. Among the three materials, PNSAC exhibited the highest adsorption capacity, which was attributed to its well-developed porosity and rich surface functional groups. In addition to the experimental work, machine learning algorithms were applied to model and predict adsorption performance. The predictive capabilities of Extreme Gradient Boosting, Random Forest, and Decision Tree algorithms were compared. Extreme Gradient Boosting achieved the highest accuracy with an R2 value of 0.97, while Random Forest and Decision Tree yielded 0.94 and 0.88, respectively. The models were validated using fivefold cross-validation and further supported by learning curves and residual distribution plots. Feature importance analysis revealed that AMX concentration and adsorbent dosage were the most influential variables affecting adsorption capacity, which was also confirmed by Pearson correlation analysis. By integrating experimental findings with interpretable machine learning models, this study offers a scalable and reliable framework for optimizing antibiotic removal from aqueous environments using low-cost and renewable biomass-based adsorbents.