Development of a multi-classification neural network model to determine the microbial growth/no growth interface

Area of Science:

• Microbiology
• Computational Biology
• Food Science

Background:

• Predicting microbial growth at limiting conditions is crucial for food safety.
• Distinguishing between growth and no growth can be challenging under these conditions.
• Existing boundary models require refinement to account for microbial behavior variability.

Purpose of the Study:

• To develop a multi-classification model for predicting microbial growth, growth transition, and no growth.
• To improve the accuracy of microbial boundary models using machine learning.
• To enhance food safety decision-making by better defining microbial growth interfaces.

Main Methods:

• Categorization of Staphylococcus aureus growth data into three classes: growth (G), growth transition (GT), and no growth (NG).
• Modeling using a Radial Basis Function Neural Network (RBFNN) combined with a memetic algorithm (MA) for oversampling.
• Development of the Smote Memetic Radial Basis Function (SMRBF) multi-classification model.

Main Results:

• The SMRBF model achieved 86.30% accuracy on training data and 82.26% on generalization data.
• The model effectively classified Staphylococcus aureus into G, GT, and NG classes.
• High replication (n=30) resulted in a smooth transition zone, with higher GT probability at stringent conditions.

Conclusions:

• The SMRBF model accurately predicts microbial growth boundaries and transition variability.
• The inclusion of a 'growth transition' class refines the prediction of microbial behavior at interfaces.
• This model offers a valuable tool for improving food safety assessments and decision-making processes.