Evaluation concerning the presence of bacteria in meat products is mandatory for commercializing these goods. Although food bacteria detection is based on microbiological methods, these assays are usually laborious and time-consuming. In this paper, an electronic nose is used to differentiate Salmonella spp. (SA), Escherichia coli (EC), and Pseudomonas fluorescens (PF) inoculated in raw meat (beef, chicken, and pork) and incubated at 22 °C for 3 days. The obtained data were evaluated by principal component analysis (PCA) and different machine learning algorithms. From the graphical analysis of the PCA, on day 1, the clusters were close to each other for beef, chicken, and pork, while on days 2 and 3, more separated bacteria clusters were obtained regardless of the meat type, allowing for the discrimination of the samples for the latter days. To estimate the growth rates of the microorganisms, the distance between clusters was calculated and provided a pattern for the three bacteria, with the slowest-, moderate-, and fastest-growing being EC, SA, and PF, respectively. Concerning the machine learning algorithms, the accuracy varied from 93.8 to 100% for beef and chicken, while for pork, it varied from 75% to 100%. Thus, these results suggest that the proposed methodology based on electronic nose has the potential for the direct discrimination of bacteria in raw meat, with reduced analysis time, costs, and manipulating steps.
Keywords: electronic nose; food safety; foodborne bacteria; machine learning; meat; microbiology.