Odor Fingerprinting of Drinking Water Using a GC-Based Electronic Nose: A Rapid Screening Framework for Nine Odorants

This study presents a gas chromatography-based electronic nose (GC-E-nose) approach for screening and classifying the response patterns of nine predefined odorants in drinking water. The PLSR models provided compound-level concentration prediction, with R2 values ranging from 0.86 to 0.99 and RMSE values below 27 µg/L. LDA effectively visualized compound- and concentration-dependent response patterns. Canonical discriminant analysis (CDA) showed moderate discrimination among the response classes of representative standards; preparation-level leave-one-out cross-validation (LOOCV) of the 225 valid standard-solution observations yielded an overall sample-level accuracy of 74.7% and a macro-averaged sensitivity of 77.6% across the six predefined response classes. Because none of the target ODCs was detected in the field samples, CDA could not be externally validated. Instead, matrix-matched spiked samples were used to evaluate PLSR, supporting the proposed method as a rapid screening tool prior to confirmatory GC-MS analysis.