Designing cost-effective monitoring schemes for chemical hazards in feed using machine learning

Chemical hazards, like mycotoxins, dioxins and polychlorinated biphenyls (PCBs), and heavy metals in feed can affect animal and human health. Maximum allowable concentrations of these food safety hazards in animal feed and their ingredients are established in EU Directive 2002/32/EC to control their presence in feed. Feed containing such food safety hazards above the maximum levels may be unsafe for animals and humans and must be withdrawn from the supply chain. Therefore, the presence of such food safety hazards in feed can result in economic damage due to the recall of the contaminated feed as well as food of animal origin. Also, economic damage occurs through higher disease prevalence in case the contaminations are not detected. Monitoring programs to control the presence of chemical hazards in feed have been designed and implemented by both the industry and governmental agencies. Since the checking the presence of all food safety hazards in the endless number of feed ingredients is resource demanding, such monitoring programs are ideally carried out with a risk-based approach to monitor the hazards that pose the highest risk to animal and human health. In addition, monitoring plans could be conducted in a cost-effective way, meaning the plan provides the highest effectiveness of food safety monitoring given available resources. ML algorithms can be used to design food safety monitoring schemes since they can learn from existing food safety-related data and other information sources to make predictions of the occurrence of food safety hazards and their related risks. Using ML for designing cost-effective monitoring plans needs a comprehensive approach, with a deeper investigation of multiple aspects, such as the ML model performance, the unbalanced nature of the food safety monitoring data, the economic losses due to incorrect model results, the estimation of monitoring effectiveness and cost, and the optimal sample size. The overall objective of this project is to design methods to design cost-effective monitoring plans for chemical hazards in feed.