Rapid and accurate identification of foodborne bacteria: a combined approach using confocal Raman micro-spectroscopy and explainable machine learning.

This study proposes a rapid identification method for foodborne pathogens by combining Raman spectroscopy with explainable machine learning. Spectral data of nine common foodborne pathogens are collected using a laser confocal Raman spectrometer, and their characteristic Raman peaks are identified and analyzed. Key spectral features are extracted using competitive adaptive reweighted sampling (CARS) and the successive projections algorithm (SPA), while t-distributed stochastic neighbor embedding (t-SNE) is employed for visualization. Subsequently, classification models, including support vector machine (SVM) and random forest (RF), are developed, and the optimal model is selected based on classification accuracy (ACC), with the RF model achieving a test accuracy of 98.91%. To enhance the interpretability of the model, Shapley Additive exPlanations (SHAP) analysis is applied to evaluate the contribution of each spectral feature to the classification results, identifying critical Raman shifts significantly influencing pathogen classification. The results demonstrate that CARS-SPA feature selection not only improves the accuracy and efficiency of the classification model but also enhances its transparency and reliability. This study optimizes the workflow for food safety testing, reduces the risk of foodborne diseases, and provides robust technical support for public health and safety.