High-capacity removal of crystal violet using ZIF-8/graphene quantum dot composite with RSM optimization and explainable machine learning.

Synthetic dyes are persistent pollutants resistant to conventional treatment, necessitating effective removal strategies. This study examines the adsorption of Crystal Violet (CV) onto a ZIF-8/graphene quantum dot (Z8GD) composite under varying conditions. Batch experiments revealed strong sensitivity to operational parameters, with capacities ranging from 76 to 971 mg/g. Adsorption capacity increased from 195 to 460 mg/g as the dose decreased (0.10 → 0.04 g/L), from 200 to 401 mg/g with higher CV concentration (25 → 75 ppm), and from 162 to 971 mg/g with longer shaking time (3 → 24 h). Response Surface Methodology identified these factors as highly significant (p < 0.0001) and yielded a robust predictive model (R² = 0.9869). Kinetic analysis showed that the Avrami model (R² = 0.9993) best described the process, suggesting multi-mechanistic uptake. The maximum adsorption capacity reached ~ 7162 mg/g, with the Redlich–Peterson isotherm providing the best fit (R² = 0.9969). Thermodynamic analysis indicated an endothermic process (ΔH = 20.9 kJ/mol), with Gibbs free energy becoming more negative at higher temperatures (ΔG = − 30.6 to − 33.9 kJ/mol). Post-adsorption XRD and FTIR confirmed Z8GD’s structural stability and revealed multiple interactions, including π–π/CH–π stacking, hydrogen bonding, and electrostatic attraction. Machine learning models further enhanced predictive capability, with the SVR + XGB hybrid achieving the highest accuracy (R² = 0.9986). Shapley Additive Explanations identified shaking time and initial dye concentration as the most influential variables. Overall, Z8GD demonstrated exceptional adsorption capacity and mechanistic versatility, while the integration of RSM and ML provided both optimization and interpretability for adsorption behavior.