A First Approach to combine AI and Predictive Microbiology

Introduction: This study analyses the application of predictive microbiology as a strategic tool for risk assessment and the optimization of food shelf-life, focusing on a first approach for the development of a unified third-level predictive model capable of estimating the behaviour of Listeria monocytogenes based on multiple environmental and technological variables.

Methods: The methodology initially involved a simulation phase conducted using two reference software programs, FSSP and PMP, based on a multivariate experimental design (DoE). Variables such as temperature, pH, NaCl concentration, nitrite, lactic acid, and initial microbial load were analyzed. The data obtained were then processed using the Gompertz equation to extract kinetic parameters, mainly t₁₀₀, which represent the time required for a 100-fold increase in the microbial population.

Results: Multivariate statistical analysis, conducted through Pareto chats and response surfaces, highlighted that temperature and pH are the primary determinants of growth rate, while nitrite exerts the most significant inhibitory effect. In the final phase, the integration of data through computational algorithms and the support of generative artificial intelligence allowed for the comparison of various regression models.

Conclusion: The non-linear logistic model was selected as the most effective due to its robustness and biological consistency, proving superior to purely mathematical models like Random Forest in its ability to avoid unrealistic extrapolations. The results confirm that an integrated approach combining predictive software and advanced statistics serves as a fundamental support for designing HACCP self-control systems and for the proactive management of food safety within the Industry 4.0 framework.