Risk predictions
for poultry pathogens
Key Objective
To make data useful for short-term risk trends and longer-term emerging risks in the poultry industry.
Despite all the advances in digital connectivity (sensor networks, IOTs etc.) creating a data-driven Risk Assessment in the poultry industry is time consuming and labour intensive. Nowadays, the volumes of data created on a daily basis-worldwide are huge (e.g. environmental data, lab tests data etc.) and heterogeneous with significant diversity and complexity, in many formats, types and languages with missing values and insufficient context.
Not taking full advantage of this wealth of public and private data comes at a great cost: despite best efforts and modern techniques, consumers world-wide still get sick from poultry diseases and related companies suffer huge economic and legal penalties from product food recalls.
This Use Case will deploy careful machine learning engineering and access to extensive sources of data (private and public) to train AI models on a variety of examples for short-term risk trends and longer-term emerging risks in the poultry industry.
- Lab test data for the presence of three particular pathogens: (a) listeria – from factory processes, (b) campylobacter – from hatchery and farm processes, (c) salmonella – from hatchery and farm processes (MOY Databases)
- Environmental data from hatchery and farm sensors, such as temperature (internal and external), air humidity, CO2 levels, light levels (MOY Databases)
- Incidents (e.g., product recalls, border rejections) for poultry products (Public Data)
- Lab test results from public authorities for poultry ingredients (Public data)
- Inspection results for poultry ingredient suppliers (Public Data)
- Food safety news and expert opinion pieces on current and emerging pathogen risks in poultry (Public Data)
- Consumer reports and discussions concerning poultry products and ingredients (Public Data)
- Relevant scientific reports and publications (Public Data)
This use-case will deploy and train appropriate AI models using the EFRA privacy-preserving two-level training approach.
At the first level, an AI model is trained on the public and private aggregated data of MOY to provide them with a personalised AI model for poultry pathogen predictions.
At the second level, EFRA will leverage the AGROKNOW and SGS Digicomply customer base with 2 additional food companies in the same or adjacent industries to train a single AI model using a federated learning approach.
SGS will also participate by providing appropriate lab test results and/or other meat/poultry food- safety-related data records. To incorporate the environmental sensor readings EFRA will develop re-usable cloudlet applications that will run on the local farm premises, pre-process the sensor readings, and deliver higher quality, consolidated results at a slower velocity, such that they can be directly mapped to the AI model parameters.
EFRA will monitor and predict the presence of the most usual pathogens in the poultry industry: listeria, campylobacter and salmonella. Lab test results and environmental data (such as temperature, air humidity, light levels) coming from farm sensors, will be provided by Moy Park databases and equipment. Combining them with publicly available data sources, related to poultry industry (such as incidents for products and inspection results) this use-case will deploy and train appropriate AI models for poultry pathogen predictions.
Latest news about the Use Case
Funded by the European Union. Views and opinions expressed are however those of the author(s) only and do not necessarily reflect those of the European Union or European Commission-EU. Neither the European Union nor the granting authority can be held responsible for them.
2023 © EFRA project