Machine-learning-driven evaluation of superparamagnetic activated carbon for continuous tetracycline removal

In this work, a magnetic activated carbon (CPL–MAC) was synthesized from fallen Copper Pod leaves and evaluated for continuous tetracycline (TC) removal in a fixed-bed column. The influence of bed height (Z = 1.0, 1.5, 2.0 cm), flow rate (Q = 1, 2, 4 mL/min) and influent concentration (C₀ = 50, 100, 150 mg/L) on breakthrough behavior, adsorption capacity and bed utilization were systematically studied at pH 4. Breakthrough curves showed typical S-shaped profiles, with higher Z extending breakthrough and exhaustion times (t_e), and higher Q and C₀ causing earlier saturation. The column displayed its best overall performance at Z = 1.5 cm, Q = 2 mL/min and C₀ = 100 mg/L, where the Thomas model predicted a maximum capacity of 271.6 mg/g. Thomas, Yoon–Nelson, Adams–Bohart and BDST models fitted the experimental data well (R² > 0.95), with BDST giving reliable service-time estimates for different breakthrough criteria. In addition, machine learning models (SVR, ANN and tree-based ensembles) were developed to predict C_t/C₀, with CatBoost giving the highest accuracy (R² = 0.998, RMSE = 0.017). Feature-importance, SHAP and PD analyses indicated that time is the dominant variable, followed by Q, influent C₀ and Z. The combined integration of fixed-bed experiments, mechanistic column modelling, and explainable machine-learning analysis establishes a robust laboratory-scale framework for evaluating CPL–MAC performance toward tetracycline removal.