Automated food safety early warning system in the dairy supply chain using machine learning

14 Traditionally, early warning systems for food safety are based on monitoring targeted food 15 safety hazards. Therefore, food safety risks are generally detected only when the problems 16 have developed too far to allow preventive measures. Successful early warning systems 17 should identify signals that precede the development of a food safety risk. Moreover, such 18 signals could be identified in factors from domains adjacent to the food supply chain, so-called 19 drivers of change and other indicators. In this study, we show for the first time, using the dairy 20 supply chain as an application case, that such drivers and indicators may indeed represent 21 signals that precede the detection of a food safety risk. Using dynamic unsupervised anomaly 22 detection models, anomalies were detected in indicator data expected by domain experts to 23 impact the development of food safety risks in milk. Detrended cross-correlation analysis was 24 used to demonstrate that anomalies in various indicators preceded reports of contaminated 25 milk. Lag times of more than 12 months were observed. Similar results were observed for the 26 6 largest milk-producing countries in Europe (i.e., Germany, France, Italy, the Netherlands, 27 Poland, and the United Kingdom). Additionally, a Bayesian network was used to identify the 28 food safety hazards associated with an anomaly for the Netherlands. 29 These results suggest that severe changes in domains adjacent to the food supply chain may 30 trigger the development of food safety problems that become visible many months later. 31 Awareness of such relationships will provide the opportunity for food producers or inspectors 32 to take timely measures to prevent food safety problems. A fully automated system for data 33 collection, processing, analysis and warning, such as that presented in this study, may further 34 support the uptake of such an approach. 35 36