Machine learning combined with HS-GC-IMS and HS-SPME-GC-MS volatile fingerprinting for rapid discrimination of Listeria monocytogenes and Cronobacter sakazakii in contaminated chicken meat.

To characterize volatile organic compound (VOC) alterations related to chicken contamination by Listeria monocytogenes and Cronobacter sakazakii, HS-GC-IMS and HS-SPME-GC-MS were integrated with machine-learning methods for discriminative VOC feature selection and sample classification. The results demonstrated that both platforms captured distinct VOC patterns associated with bacterial contamination, providing complementary information for rapid fingerprint profiling and chemical characterization. Machine learning-based feature selection using recursive feature elimination, logistic regression, random forest, combined with fold-wise leave-one-out cross-validation, identified VOC features that contributed to discrimination among the investigated contamination conditions. HS-GC-IMS achieved 92.86-100% accuracy in binary classification tasks and 94.44% accuracy in three-class classification, whereas HS-SPME-GC-MS achieved 96.67-100% and 98.61% accuracy, respectively. Cyclohexanone showed a stronger association with Cronobacter sakazakii contamination, whereas e-2-hexenal, 4-methyldecane, 2-pentanone D, and 2-ethyl-1-hexanol were more representative of Listeria monocytogenes contamination. In addition, 1-octen-3-ol and ethyl acetate were more indicative of general contamination- or spoilage-related changes rather than pathogen-specific responses. Dynamic analysis during 0-24 h highlighted undecane, cyclohexanone, 2-ethyl-1-hexanol, 4-methyldecane, e-2-hexenal, and 1-octen-3-ol throughout the initial contamination phase, indicating their potential value for earlier contamination screening. Overall, the integration of VOC fingerprinting with machine learning provides a promising strategy for rapid and interpretable discrimination of pathogen-inoculated chicken samples under controlled experimental conditions.