Initializing a Public Repository for Hosting Benchmark Datasets to Facilitate Machine Learning Model Development in...

Area of Science:

• Food Safety
• Artificial Intelligence
• Machine Learning
• Data Science

Background:

• Development and deployment of AI/ML in food safety are hindered by a lack of accessible, well-curated datasets.
• Publicly available databases are crucial for developing and validating AI/ML models in this domain.

Purpose of the Study:

• To establish a public repository of curated food safety datasets to facilitate AI/ML model development.
• To provide benchmark datasets and tools for training and validating predictive models for foodborne pathogens.

Main Methods:

• Consolidated and curated three existing datasets on Listeria, Salmonella, Campylobacter, and E. coli contamination.
• Made datasets publicly available in the Cornell Food Safety ML Repository.
• Developed customizable and screening scripts (LazyPredict) for training various ML models.

Main Results:

• The Cornell Food Safety ML Repository now hosts curated datasets covering soil, poultry, and watershed contamination.
• Demonstrated the utility of the repository with scripts for ML model training.
• Highlighted the need for continued data contributions to advance predictive food safety modeling.

Conclusions:

• The Cornell Food Safety ML Repository is a vital resource for AI/ML in food safety, addressing data accessibility challenges.
• Encourages further contributions of well-curated datasets to expand predictive modeling capabilities.
• Discusses the benefits of public databases and privacy-preserving data-sharing techniques.