Machine learning for optimization of zeolites for boric acid sorption

Machine learning can effectively accelerate materials development in real‐world sorbent applications including clean‐up of polluted impoundment sites. Zeolites synthesized from coal fly ash can adsorb contaminants, such as boric acid, from water. Machine learning models were trained on molecular simulation data to predict boric acid uptake based on zeolite structure, aluminum content, extra‐framework cation species, and boron concentration in solution. Overall, eXtreme Gradient Boosting models yielded the highest speed and accuracy. The models were used with a genetic algorithm to enable concentration‐specific zeolite optimization for coal ash impoundment sites. Results indicate small pore zeolite frameworks such as PHI, CHA, AVE, ERI with low Si/Al ratios and a mix of Na and Ca metal cations are most effective for boric acid removal. Our use of machine learning models with a genetic algorithm has broad implications for machine learning‐aided materials design.