Green analytical method for COD determination using UV–Vis spectroscopy combined with machine learning

The current study describes the development of a simple, direct, low-cost, and high-throughput method to determine chemical oxygen demand (COD) in wastewater samples using UV–Vis spectroscopy combined with machine learning techniques. Robust models were developed using 289 wastewater samples collected in seven distinct sewage treatment plants (STPs) in Paraná state, Brazil. Principal components analysis (PCA) showed a considerable tendency to group samples from STPs that employed similar wastewater treatments. Multivariate calibration models were built using partial least squares (PLS) regression combined with inspection of regression vector and ordered predictors selection (OPS) to select relevant spectral variables. Results indicate the models are suitable to predict COD, with root mean square error of calibration (RMSEC) and prediction (RMSEP) lower than 14.19 and 15.00 mg L⁻¹ O₂, respectively. Additionally, the models showed ratio performance deviation (RPD) higher than 2.75, indicating an excellent prediction accuracy. The analytical GREEnness (AGREE) score for the proposed method is 0.82, confirming its greenness characteristic. These results demonstrate that the proposed method can be applied in COD determination, allowing fast sample screening at a low cost with no solvent consumption and generation of waste.