Portable Metal Oxide E-Nose Based on Machine Learning for Multiple Beverage Classification

Beverage quality control is crucial for ensuring consumer safety, product authenticity, and market integrity. Traditional methods such as high-performance liquid chromatography and mass spectrometry are effective but present limitations in terms of cost, complexity, and time efficiency. In this study, we propose an electronic nose (E-nose) platform, combined with a linear discriminant analysis (LDA) algorithm, as a novel approach for beverage classification. The E-nose system, comprising a gas sensor array, was evaluated using multiple beverage samples, including alcoholic beverages, coffee, tea, and soft drinks. Key features were extracted from the sensor responses, and classification tasks were performed using machine learning techniques. Our results demonstrate that the LDA classifier achieves a high inter-class classification accuracy, with an AUC of $\mathbf{1. 0 0}$ for most beverage categories. However, intra-class classification, particularly among alcoholic beverages, remains challenging, with AUC values dropping to 0.64 for rosé wine. While the overall intra-class accuracy was $87.5 \%$, further optimization is needed to improve discrimination between beverages with similar volatile profiles. These findings highlight the potential of E-nose technology for rapid, cost-effective beverage quality control, with opportunities for future enhancements through advanced sensor arrays and machine learning models.