Investigating the removal of Congo red dye using ZIF-8 and GQD composite: Characterization, kinetics, isotherm, thermodynamics, optimization, and machine learning studies

This study explores the efficacy of zeolitic imidazole framework-8 (ZIF-8) and graphene quantum dots (GQDs) composite (abbreviated as Z8D) in eliminating Congo Red (CR) dye from aqueous solutions. Material characterization using FESEM and TEM revealed well-dispersed GQDs on ZIF-8 cubes, confirming composite synthesis. Response surface methodology (RSM) was utilized to examine the effects of adsorbent dosage, initial CR concentration, and contact time on adsorption capacity, providing useful insights into the adsorption process. Specifically, increasing the initial CR concentration and contact time led to an increase in the CR adsorption capacity from 96.5 to 164.9 mg/g and from 176 to 267.4 mg/g, respectively, for definite operating conditions. Additionally, decreasing the adsorbent dose from 100 to 40 mg/L boosted the CR adsorption capacity by about 50 % (i.e., from 164.9 to 246 mg/g). Furthermore, multiple adsorption models were successfully combined using advanced machine learning (ML) models— Extreme Gradient Boosting (XGB), Support Vector Regression (SVR), Random Forest (RF)—and their ensemble combinations (SVR-RF, SVR-XGB, XGB-RF) to predict adsorption performance with high precision. All ensemble combinations yielded higher test R² values and lower prediction errors compared to their corresponding individual base models. Shapley Additive Explanations (SHAP) analysis was applied to unravel the black-box nature of these models, identifying contact time as the most crucial factor, followed by initial CR concentration and adsorbent dosage. The adsorption kinetics followed the Avrami model (R² = 0.9999), suggesting a complex mechanism. Additionally, isotherm studies demonstrated exceptional fits to the Langmuir (R² = 0.9966) and Redlich-Peterson (R² = 0.9973) models, with a maximum Langmuir adsorption capacity of 421.3 mg/g. Thermodynamic analysis confirmed the endothermic and spontaneous nature of CR adsorption (∆H = 7.98 kJ.mol⁻¹, ∆S = 107.6 J.mol⁻¹·K⁻¹). The XGB-RF ensemble emerged as the best ML model (R² = 0.9441), underscoring its predictive capability. This work establishes Z8D as a promising adsorbent for CR removal from wastewater and highlights the transformative role of ML in advancing adsorption science.