Machine learning modeling of electrochemical ammonia removal using an integrated dataset: Model simplification and conceptual application

Electrochemical treatment emerges as an attractive approach for ammonia removal, which is essential for sustainable wastewater treatment. To identify key determinants during electrochemical ammonia removal and improve predictive capability, a dataset of 517 experimental observations from 40 studies was compiled and analyzed using machine learning. Among six algorithms, Random Forest (RF) and Extreme Gradient Boosting (XGBoost) achieved the highest predictive accuracy, with test-set R 2 of 0.71–0.73 and RMSE of 15.8–15.9%. Model interpretation identified total ammoniacal nitrogen (TAN) as the dominant inhibitory factor, reducing removal efficiency from ∼70% at <0.05 g TAN L −1 to <50% above 4.5 g TAN L −1 . Total chemical oxygen demand (TCOD) and hydraulic retention time (HRT) exerted strong but nonlinear influences, while electrical load value 1 had some secondary effects. Simplified models using only TAN, TCOD, HRT, and electrical load value 1 retained most predictive capacity (R 2 = 0.69–0.71), providing a concise and interpretable framework. A conceptual application of the simplified models with real wastewater data showed that influent TAN and TCOD significantly constrained achievable removal efficiency (∼60–68%), while substantial improvements (>90%) would require influent dilution. These findings have quantitatively defined operational limits of electrochemical ammonia removal and revealed the potential of machine learning to support system optimization and future scale-up. • Machine learning models are developed to simulate electrochemical ammonia removal. • An integrated dataset of 517 experimental observations from 40 studies is analzyed • Random Forest and XGBoost achieved high predictive accuracy. • Simplified models with four key features maintain predictive capacity and improved interpretability. • A conceptual application using real wastewater data achieves high removal efficiency.