Robust prediction of Listeria monocytogenes, Listeria innocua and aerobic spoilage bacteria growth in food systems using Multilayer Perceptron Artificial Neural Networks

Spoilage microorganisms and pathogenic bacteria can negatively impact food safety and stability, leading to potential health risks and economic losses. Predictive microbiology is a crucial tool for forecasting microbial behavior in food matrices, and supervised machine learning stands out as an effective solution for handling complex datasets (e.g., multiple variables and non-linear relationships). This work aims to develop a growth prediction model for Listeria monocytogenes, Listeria innocua and aerobic spoilage bacteria in food and culture media using artificial neural networks. The model development involved the implementation of the Multilayer Perceptron algorithm applied to datasets obtained from the ComBase platform. We evaluated different models to understand the impact of dataset size, additional input variables, food matrix data, and microorganism data on model performance. Experimental data were also used for model validation. Results demonstrate favorable statistical metrics for all models ( R 2 = [0.847 - 0.929], R M S E = [0.55 - 0.852], B f = [1.002 -1.011] and A f = [1.071 - 1.108]), indicating the efficacy of the proposed approach for the intended application. Furthermore, validation of the model using experimental data on the growth of total aerobic spoilage bacteria in beef demonstrated its ability to accurately describe the kinetics of microbial growth. These findings underscore the potential of Multilayer Perceptron models to predict microbial growth in various food matrices and culture media, exhibiting robust performance and generalization capability. • Multilayer Perceptron accurately predicted bacterial growth across diverse conditions. • Growth bacteria models were developed using ComBase platform data. • MLP neural networks outperforming XGB in predicting Listeria growth. • Bacteria growth model performed remarkably well during independent data validation. • Model was robust in predicting bacterial growth across several food systems.