Amoxicillin removal by babassu-derived activated carbon under ultrasonic irradiation: adsorption performance and tree-based machine learning modeling

Abstract The contamination of aquatic environments by antibiotics has become a major environmental concern due to their persistence and contribution to antimicrobial resistance. In this study, activated carbon derived from babassu endocarp (BAC) was evaluated for the removal of amoxicillin (AMX) from aqueous solutions under ultrasonic irradiation. BAC exhibited a heterogeneous porous structure with a specific surface area of 623.38 m 2 g −1 . Adsorption equilibrium was achieved within approximately 10 min under ultrasonic irradiation. This equilibrium time is shorter than equilibrium times reported for many activated carbons used for AMX removal. The Freundlich model provided the best fit to the equilibrium data, while the Langmuir model predicted a maximum adsorption capacity of 185.57 mg g −1 . Thermodynamic analysis indicated that the adsorption process was spontaneous and endothermic. BAC also maintained satisfactory adsorption performance after four regeneration cycles. In addition, tree-based machine learning algorithms, including Decision Tree (DT), Random Forest (RF), and Extra Trees (ET), were applied to predict adsorption capacity, with the ET model achieving the best predictive performance ( R 2 = 0.983) in the test dataset. Feature importance analysis identified the initial AMX concentration as the most influential operational variable. The adsorption mechanism was proposed to involve pore filling, hydrogen bonding, π–π interactions, and electrostatic attractions. Overall, the results demonstrate that BAC is a promising low-cost adsorbent for rapid AMX removal and highlight the applicability of machine-learning approaches for adsorption process prediction.