Machine learning–guided prediction and optimization of cyanide removal using NaClO/ClO2 dual-oxidant alkaline chlorination in metal-plating wastewater

Cyanide-containing wastewater generated from metal-plating industries poses a serious environmental threat owing to the high toxicity and persistence of metal–cyanide complexes. In this study, cyanide removal using a dual-oxidant NaClO/ClO 2 alkaline chlorination system was systematically investigated and optimized using a machine learning (ML)–assisted experimental framework. Initial experiments were conducted using synthetic wastewater containing Ni–, Zn–, and Fe–cyanide complexes to assess the effects of key operational parameters. Tree-based ML models were trained on the experimental data to identify influential variables and estimate effective oxidant dosage ranges. Based on the model predictions, a central composite design (CCD) was applied to optimize the total chlorine dose and NaClO:ClO 2 ratio for real metal-plating wastewater. The CCD results were subsequently analyzed using an ensemble of regression models with diverse learning architectures. The optimal operating conditions were determined by aggregating model-specific optima and applying density-based clustering to identify the consensus regions. The ensemble-derived optimum corresponded to a total chlorine dose of 84.5 mg/L and a NaClO:ClO 2 ratio of 1.5:8.5. Experimental validation under these conditions achieved a cyanide removal efficiency of 79.2%, which was consistent with the ensemble predictions. These results demonstrate that ensemble-based ML approaches offer a robust and reliable strategy for optimizing complex oxidation processes in industrial wastewater treatment.