Integrating statistical and machine learning approaches to characterise and model the performance of a full-scale wastewater treatment plant

), capturing non-linear and regime-dependent behaviours and outperforming simpler models. GA optimised pollutant-specific ANN structures, improving both accuracy and interpretability. The models consistently identified influent chemical oxygen demand, ammonium-nitrogen and aeration as key variables controlling pollutant removal. Organic matter removal was strongly influenced by the balance between influent composition, aeration and hydraulic conditions. Distinct operational regimes were identified, with influent suspended solids, aeration and flow rate governing nitrogen, phosphorus and chemical oxygen demand dynamics, respectively. Thus, the proposed framework enabled the identification of key drivers and operational thresholds influencing effluent quality, supporting improved process understanding and optimisation of wastewater treatment processes.