Traceability analysis of tobacco products based on GC-MS and electronic nose detection technology

Abstract To achieve cross-regional precise identification of tobacco varieties and explore efficient, reliable, and low-cost methods for origin traceability, this study establishes a tobacco traceability framework by integrating gas chromatography-mass spectrometry (GC-MS) technology with machine learning algorithms. First, GC-MS was employed for qualitative and quantitative analysis of volatile components in tobacco, guiding the selection of high-sensitivity commercial electronic nose sensors to construct a tobacco detection system. Subsequently, the collected data underwent preprocessing and feature extraction, followed by classification and prediction using five algorithms: Support Vector Machine (SVM), Random Forest (RF), Back-Propagation Neural Network (BP), Radial Basis Function Neural Network (RBF), and Convolutional Neural Networks (CNN). Experimental results demonstrated that the STB (extracted value) feature significantly outperforms the MAX (maximum value) and STB-MAX (hybrid parameter) features. Both SVM and CNN achieve the highest classification accuracy of 98.28% under the STB feature, which is 5.18% and 10.35% higher than the results under the STB-MAX feature (93.10% and 87.93%) respectively, and significantly outperform other algorithms (RF: 93.10%, BP: 96.55%, RBF: 94.83%). In terms of result variability, SVM has the lowest variability (dispersion of approximately 5%) due to its structural risk minimization mechanism; RF and CNN are sensitive to feature quality; and RBF and BP have stability levels between the former two. This difference confirms the critical impact of the match between feature robustness and algorithm adaptability on classification results. By optimizing feature selection and algorithm adaptation mechanisms, this study provides an efficient solution for tobacco origin traceability. The proposed methodological framework can be extended to quality control scenarios in the food and pharmaceutical industries.