Developing machine learning models for fluid milk spoilage classification.

Area of Science:

• Food Science
• Microbiology
• Artificial Intelligence

Background:

• The dairy industry faces knowledge gaps due to expert retirement.
• Identifying fluid milk spoilage patterns is crucial for effective control strategies.
• Traditional spoilage classification relies on human experts.

Purpose of the Study:

• To develop a machine-learning-based digital expert system for classifying fluid milk spoilage.
• To assess the potential for optimizing microbiological testing schemes in the dairy industry.

Main Methods:

• A machine-learning model was developed using microbiological data from 770 fluid milk samples.
• Expert-assigned spoilage types (Gram-negative bacteria, sporeformers, no spoilage) were used for training and validation.
• Multiple models were trained and tested using subsets of data representing optimized testing scenarios.

Main Results:

• The baseline model achieved 96.4% classification accuracy on the test set.
• Optimized models using reduced data sets (e.g., specific microbial counts on day 14 and 21) reached 94.2% testing accuracy.
• The developed system can identify predominant spoilage patterns and aid in root-cause investigations.

Conclusions:

• Machine-learning models offer a viable solution for classifying fluid milk spoilage.
• Optimized testing schemes can reduce costs and resources while maintaining high accuracy.
• This digital expert system can support targeted interventions and improve quality control in the dairy sector.