Food supply chains are complex and vulnerable to many factors (e.g. climate, economy and human behaviour) having a direct and/or indirect effect on the development of food safety risks and/or food fraud. All of these factors should be taken into account in approaches aiming to predict food safety problems or food fraud at an early stage. Hence, a system approach is needed that takes into account this web of interactions between influencing factors and that can utilise the huge amount of data of various sources and nature. This will enable the identification of a food safety risk or food fraud at its early stage of development and allow timely mitigation actions to prevent the risk to occur.
Solution / approach
The ambition of this project was to apply machine learning, in particularly Bayesian Network (BN) analysis in food safety to generate more and better knowledge from the huge amount of data being generated in house but also elsewhere in the world.
This data can be open source data ( such as economic, agricultural, climate data, etc.), or confidential monitoring data. Using fruit and vegetable and the dairy supply chain as showcases, it was demonstrated that BN models could predict the hazard type and/ or the contamination level with more than 90% accuracy. This approach is now being evaluated for implementation by different stakeholders (industry, authorities).
Advanced analytics were implemented on a shared data infrastructure (i.e. WUR HPC) that is adapted to facilitate sharing of data, Machine Learning algorithms and knowledge while ensuring the ownership and privacy of sensitive data. Models and data sources are made accessible for the user via this infrastructure (see figure).