Machine learning modelling and mechanistic insights into malachite green adsorption on sustainable magnetic activated carbon

This study presents a sustainable route for transforming Spathodea campanulata pods into pod magnetic activated carbon (PMAC) for efficient removal of Malachite Green (MG) from water. The sustainability of this approach stems from the use of freely available biomass, mild low-temperature carbonization, inexpensive precursors, and magnetic recoverability that minimizes energy and chemical demands during post-treatment. PMAC had a high specific surface area (877.07 m 2 /g) with a mesoporous structure (pore size: 2.51 nm). Comprehensive characterization revealed a rough, porous surface enriched with functional groups that drive adsorption through π–π stacking, electrostatic attraction, hydrogen bonding, and surface complexation. At the most favorable operational conditions (pH 7.0, 0.15 g/L PMAC, 100 mg/L MG, 303 K), the prepared PMAC demonstrated a Langmuir adsorption capacity of 783.47 mg/g. The kinetic trend followed the pseudo-second order model, while the equilibrium data agreed better with the Freundlich isotherm. From the thermodynamic evaluation, the process was spontaneous, endothermic, and entropy driven. In the practical test with industrial groundwater, PMAC delivered 492.44 mg/g, and after six methanol regeneration cycles, it still held over 77 % of the initial performance. For modelling, the Adaptive Neuro-Fuzzy Inference System (ANFIS) tracked the data very closely (R 2 = 0.9953, MSE = 0.00098, MAE = 0.0227), and it outperformed artificial neural networks. The variables ranked by influence were initial concentration > contact time > dosage > pH > temperature. Taken together, the experiments and the modelling show that PMAC is a reusable and scalable option for removing MG from complex wastewater. • Spathodea pod-derived PMAC adsorbs Malachite Green up to 783.47 mg/g • Robust adsorption confirmed by Freundlich isotherm and PSO kinetics • Reusable adsorbent retains >77 % capacity after six methanol cycles • Real-water spiking shows 73.86 % MG removal in industrial groundwater • ANFIS model gives R 2 = 0.9953 and lowest prediction errors among ML tools