Novel insight into the heavy metal immobilization by coal fly ash-based geopolymers using machine learning and density functional theory

Coal fly ash (CFA)-based geopolymers are sustainable low-carbon binders for heavy metal immobilization, while promoting solid waste utilization and safe disposal. CFA-based geopolymers immobilization of heavy metal primarily depends on the raw material properties, curing conditions, alkali activator properties and heavy metal properties. However, conventional methods for optimizing geopolymer synthesis and evaluating immobilization capacity are costly, time-intensive, and lack of insight into solidification mechanisms. This study combined machine learning (ML) algorithm and density functional theory (DFT) to predict and reveal heavy metal immobilization of CFA-based geopolymers. Eight ML models were evaluated, with the gradient boosting regression (GB) model exhibiting the best predictive performance (R² = 0.9284, RMSE = 0.3912). Feature importance analysis reveals determinants of immobilization performance: heavy metal properties > geopolymer raw material properties > curing conditions > alkali activator properties. DFT calculations revealed that geopolymers incorporating large-radius hydrated heavy metal ions, low Si/Al ratios, and elevated calcium content exhibit enhanced heavy metal immobilization capacity, characterized by reduced interaction energies and stronger electron localization function peaks. Overall, the integrated ML + DFT method improves predictive capabilities for complex waste systems and reveals immobilization mechanisms.