Rapid differentiation of water faecal indicators from coliforms using redox potential curves and machine learning

Abstract Accurate and rapid identification of waterborne faecal indicators, such as Escherichia coli and Enterococcus faecalis , is critical for ensuring water safety. This study evaluated the use of bacterial redox potential curves as a robust approach to discriminate faecal indicators from other coliforms ( Citrobacter freundii , Enterobacter aerogenes , Klebsiella oxytoca ). Three machine learning classifiers, Support Vector Machines (SVM), Linear Discriminant Analysis (LDA), and Quadratic Discriminant Analysis (QDA), were trained on both first-order parameters and Gompertz-derived parameters. Across all classifiers, first-order parameters consistently outperformed Gompertz-based models. SVM achieved 94.5% ± 1.59% accuracy (Kappa = 0.929) with first-order parameters, compared with 89.1% ± 2.13% (Kappa = 0.860) for Gompertz inputs. LDA and QDA showed similar trends, with accuracy improvements of 13.5 and 12.7%, respectively, when first-order parameters were used. A binary classification demonstrated near-perfect discrimination of E. coli , with SVM reaching 99.1 ± 0.68% accuracy (Kappa = 0.978) using first-order inputs. Class-level analysis further confirmed excellent performance for indicator organisms, with F1-scores of 0.983 for E. coli and 0.984 for E. faecalis . These results suggest that first-order redox potential parameters effectively capture the unique metabolic characteristics of each bacterial species, allowing for accurate and consistent discrimination of faecal indicators.