Search

The massive rise of Big Data generated from smartphones, social media, Internet of Things (IoT), and multimedia, has produced an overwhelming flow of data in either structured or unstructured format. Big Data technologies are being developed and implemented in the food supply chain that gather and analyse these data. Such technologies demand new approaches in data collection, storage, processing and knowledge extraction. In this article, an overview of the recent developments in Big Data applications in food safety are presented. This review shows that the use of Big Data in food safety remains in its infancy but it is influencing the entire food supply chain. Big Data analysis is used to provide predictive insights in several steps in the food supply chain, support supply chain actors in taking real time decisions, and design the monitoring and sampling strategies. Lastly, the main research challenges that require research efforts are introduced.

Machine learning (ML) has proven to be a useful technology for data analysis and modeling in a wide variety of domains, including food science and engineering. The use of ML models for the monitoring and prediction of food safety is growing in recent years. Currently, several studies have reviewed ML applications on foodborne disease and deep learning applications on food. This article presents a literature review on ML applications for monitoring and predicting food safety. The paper summarizes and categorizes ML applications in this domain, categorizes and discusses data types used for ML modeling, and provides suggestions for data sources and input variables for future ML applications. The review is based on three scientific literature databases: Scopus, CAB Abstracts, and IEEE. It includes studies that were published in English in the period from January 1, 2011 to April 1, 2021. Results show that most studies applied Bayesian networks, Neural networks, or Support vector machines. Of the various ML models reviewed, all relevant studies showed high prediction accuracy by the validation process. Based on the ML applications, this article identifies several avenues for future studies applying ML models for the monitoring and prediction of food safety, in addition to providing suggestions for data sources and input variables.

The world market for food supplements is large and is driven by the claims of these products to, for example, treat obesity, increase focus and alertness, decrease appetite, decrease the need for sleep or reduce impulsivity. The use of illegal compounds in food supplements is a continuous threat, certainly because these compounds and products have not been tested for safety by competent authorities. It is therefore of the utmost importance for the competent authorities to know when new products are being marketed and to warn users against potential health risks. In this study, an approach is presented to detect new and unknown stimulants in food supplements using machine learning. More than 20 new stimulants were identified from two different data sources, namely scientific literature applying word embedding on > 2 million abstracts and articles from formal and social media on the world wide web using text mining. The results show that the developed approach may be suitable to detect “unknowns” in the emerging risk identification activities performed by the competent authorities, which is currently a major hurdle.

Identification of emerging risks in the food chain is essential if EFSA is to anticipate future needs in risk assessment, in relation to both data and methodology. The objectives and research proposed in the DEMETER project were specifically designed to support current (and future) EFSA procedures for emerging issue and risks identification by providing community resourcesto allow EFSA and EU Member State authorities to share data, knowledge and methods on emerging risks identification in a rapid and effective manner through a digital platform. To this end, an “Emerging Risk Knowledge Exchange Platform (ERKEP)” was developed as a prototype technical solution. Its design is based on a consultation on end‐users needs and the analysis of existing knowledge sharing solutions. ERKEP consists of three main components: 1) A content management system (CMS) providing the end‐user's “entry point” and Graphical User Interface (GUI) to ERKEP; 2) A web‐based data analytics platform (DAP) for sharing and executing data analytics workflows (DAWs), based on the KNIME Server infrastructure; 3) External web‐based services hosted by 3rd party service providers. Different DAWs were developed and added to the platform, these are: 1) Emerging risk identification system for the milk supply chain based on automated data retrieval; 2) NewsRadar; 3)Trending topics in news based on text mining and network analysis, and;4) Patent network analysis. Methodologies were identified to integrate social science information and data, into the emerging risk identification framework. Systematic reviews of the literature wereconducted in the areas of expert elicitation, citizen science, and behavioural science and a framework to incorporate data from Citizen Science into the EKREP platform was proposed. Finally, sustainability and maintenance of the project's outputs were conceptualized to enable use thereof beyond project DEMETER.

Background In the last decades, food produced by aquaculture has seen an impressive increase worldwide but maintaining high quality and safety is increasingly becoming a concern. It is apparent that changes in- and outside the aquaculture supply chain may act as driving forces for the introduction of food safety hazards. Knowledge on these drivers of change and their impact in the various steps in the food supply chain may help food producers to mitigate to potential risks and maintain high-quality food. Scope and approach In this study, we analysed the use of expert driven methodologies to assess and predict the effect of drivers of change on selected food/feed safety vulnerabilities in the salmon aquaculture supply chain of Norway. The presented overview is based on the findings of the “Aquarius” project, which was funded by the European Food Safety Authority (EFSA). Key findings and conclusions In this study, over 100 experts were involved and various expert elicitation methods were applied such as on-line questionnaires, interviews, Delphi and Failure Mode and Effect Analysis (FMEA). This approach resulted in a comprehensive overview of the Norwegian salmon supply chain. For each step in the supply chain, vulnerabilities for human and animal health were identified, which were prioritised by FMEA. For the two highest-ranked vulnerabilities in each step of the supply chain, drivers were identified and prioritised by expert elicitation in a Delphi study. Also, indicators and linked data sources were obtained for the highest-ranked drivers. The comprehensive information collected was integrated in a Bayesian Network (BN) model that links data sources for indicators and drivers of change. The applicability of the BN model was demonstrated for salmon health for four vulnerabilities and three steps in Atlantic salmon aquaculture. The accuracy of developed model was 81%.

International audience; Several tools were developed during the past decades to support farmers in nutrient management and to meet legal requirements such as the farm specific excretion tool. This tool is used by dairy farmers to estimate the farm specific nitrogen (N) excretion of their animals, which is calculated from farm specific data and some normative values. Some variables, like intake of grazed grass or roughage, are hard to measure. A data driven approach could help finding structures in data, and identifying key factors determining N excretion. The aim of this study was to benchmark machine learning methods such as Bayesian Network (BN) and boosted regression trees (BRT) in predicting N excretion, and to assess how sensitive both approaches are on the absence of hard-to-measure input variables. Data were collected from 25 Dutch dairy farms. In the period 2006–2018, detailed recordings of N intake and output were made during 6–10 weeks distributed over each year. Variables included milk production, feed intake and their composition. Calculated N excretion was categorized as low, medium, and high, with limits of 300 and 450 g/day/animal. Accuracy of prediction of the farm specific N excretion, and distinguishing the low and high cases from the medium ones, was slightly better with BRT than with BN. Leaving out information on intake during grazing did not negatively influence validation performance of both models, which opens opportunities to diminish data collection efforts on this aspect. Further analyses are required to confirm these results, such as cross-validation.

In this study, we aimed to demonstrate the aptness of a system approach to predict the level of contamination in a given agricultural product. As a showcase, the impact of climate and agrichemical use on the occurrence of food safety hazards in feed of dairy cows in the Netherlands was used. Data on chemical hazards in dairy cows' feed in the Netherlands for the years 2000 to 2013 were retrieved from the Dutch monitoring program KAP (Quality Program for Agricultural Products). Climate data (17 variables) and agrichemical usage figs. (6 variables) for the Netherlands were obtained from the NOAA's National Centers for Environmental Information, the European Commission Joint Research Center's Agri4Cast database, and FAO's FAOSTAT. A Bayesian Network (BN) was constructed with this data and optimized for the prediction of the contamination level. The overall accuracy of prediction of the level of contamination in feed was 90.3%. Sensitivity analysis demonstrated that many climate and agrichemical variables contributed to the prediction; however, their individual contribution was generally small. The applicability of the BN was demonstrated in more detail for grass and maize as feed components. The observed trends in contamination of these crops were accounted for by climate and agrichemical variables, with the impact varying amongst the specific variables and commodities. The variables with the highest impact were “days of precipitations in a month with ≥ 2.5 mm” and “annual use of herbicides". The results demonstrate that data-driven BNs can capture complex interactions, thereby enabling high-accuracy predictions. Whilst the applicability of this approach to the safety of dairy cows' feed in the Netherlands has thus been demonstrated, it can also be applied to other areas of food safety when a systems approach is needed. Such models can support risk assessors and risk managers in their understanding of the impacts of a given factor on food and feed safety, and inform the latter's decisions to mitigate potential risks.

The poultry meat supply chain is complex and therefore vulnerable to many potential contaminations that may occur. To ensure a safe product for the consumer, an efficient traceability system is required that enables a quick and efficient identification of the potential sources of contamination and proper implementation of mitigation actions. In this study, we explored the use of graph theory to construct a food supply chain network for the broiler meat supply chain in the Netherlands and tested it as a traceability system. To build the graph, we first identified the main actors in the supply chain such as broiler breeder farms, broiler farms, slaughterhouses, processors, and retailers. The capacity data of each supply chain actor, represented by its production or trade volumes, were gathered from various sources. The trade relationships between the supply chain actors were collected and the missing relationships were estimated using the gravity model. Once the network was modeled, we computed degree centrality and betweenness centrality to identify critical nodes in the network. In addition, we computed trade density to get insight into the complexity of sub-networks. We identified the critical nodes at each stage of the Dutch broiler meat supply chain and verified our results with a domain expert of the Dutch poultry industry and literature. The results showed that processors with own slaughtering facility were the most critical points in the broiler meat supply chain.

This report details the activities carried out in 2019 in the KB DDHT project 8 “Exploring the potential of non-destructive and non-invasive sensor technologies in food supply chains”. It compiles all four deliverables of this project as a single narrative. There are five case studies at the core of this project which address animal welfare, indoor farming, inspection of fresh food products, personalized nutrition advise and food fraud detection. Multiple activities were planned in these case studies with main focus on sensor technologies for non-destructive and non-invasive sensing. One of the key activities was to conduct a survey of sensor technologies applicable and potentially suitable for each case study. In addition, for each case study, one or more experiments were conducted which involved data acquisition using novel sensors as well as building knowledge on data analysis of data from the new sensors. Furthermore, as one of the goal of this project is to explore new sensor technologies, it is expected that this project will generate valuable data. Therefore, focus was also on FAIR data sharing and management. A workshop was organized in collaboration with WDCC on this topic. Most of the case studies also aligned with OnePlanet Research centre, and as a result multiple experiments are in planning for 2020 in cooperation with OnePlanet.

Knowing which products and hazards to monitor along the food supply chain is crucial for ensuring food safety. In this study, we developed a model to predict which types of herbs and spices products and food safety hazards should preferentially be monitored at each level of the supply chain (suppliers, border inspection points, market and consumers). A Bayesian Network method was used to develop a model based on notifications reported in the Rapid Alert System for Food and Feed and the database of the Dutch national monitoring program for chemical contaminants in food and feed over the period 2005-2014. The model was constructed by randomly selecting ca. 80% of the 3126 data records and validated using the remaining ca. 20% of the records. Model validation showed that the prediction accuracy was higher than 85%. Results showed that the sampling plan is closely related to the place where the products are checked along the supply chain, the products and the country of origin. Our approach of integrating different data sources and considering the entire supply chain can support industry and authorities at border inspection points and at all control points along the herbs and spices supply chain in setting priorities for their monitoring program.

In this study, a Bayesian Network (BN) was developed for the prediction of the hazard potential and biological effects with the focus on metal- and metal-oxide nanomaterials to support human health risk assessment. The developed BN captures the (inter) relationships between the exposure route, the nanomaterials physicochemical properties and the ultimate biological effects in a holistic manner and was based on international expert consultation and the scientific literature (e.g., in vitro/in vivo data). The BN was validated with independent data extracted from published studies and the accuracy of the prediction of the nanomaterials hazard potential was 72% and for the biological effect 71%, respectively. The application of the BN is shown with scenario studies for TiO2, SiO2, Ag, CeO2, ZnO nanomaterials. It is demonstrated that the BN may be used by different stakeholders at several stages in the risk assessment to predict certain properties of a nanomaterials of which little information is available or to prioritize nanomaterials for further screening.

Background Internet of Things (IoT) is growing exponentially and can become an enormous source of information. IoT has provided new opportunities in different domains but also challenges are apparent that must be addressed. Little attention has been paid to the potential use of IoT in the food safety domain and therefore the aim of this study was to fill this gap. Scope and approach This paper reviews the use of IoT technology in food safety. A literature review was conducted using academic documents written in English language and published in peer-reviewed scientific journals. The relevant articles were analysed using the bibliometric networks to investigate the relationships between authors, countries, and content. Key findings and conclusions IoT in food safety is a relatively new approach; the first article appeared in 2011 and has increased since then. Majority of these studies were performed by Chinese universities and the main IoT applications reported were on food supply chains to trace food products, followed by monitoring of food safety and quality. The vast majority of publications were related to food, meat, cold chain products and agricultural products. These studies used sensors to monitor mainly temperature, humidity, and location. The most frequently used communication technologies were Internet, radio frequency identifications (RFID) and wireless sensor networks (WSN). This article identifies knowledge gaps to inform the community, industry, government authorities about research directions for IoT in food safety.

The presence and development of many food safety risks are driven by factors within and outside the food supply chain, such as climate, economy and human behaviour. The interactions between these factors and the supply chain are complex and a system or holistic approach is needed to reveal cause-effect relationships and to be able to perform effective mitigation actions to minimise food safety risks. In this study, we demonstrate the potential of the Bayesian Network (BN) approach to identify and quantify the strength of relationships and interactions between the presence of food safety hazards as reported in Rapid Alert System for Food and Feed (RASFF) for fruits and vegetables on one hand, and climatic factors, economic and agronomic data on the other. To this end, all food safety notifications in RASFF (i.e. 3781 notifications) on fruits and vegetables originating from India, Turkey and the Netherlands were collected for the period 2005–2015. In addition, climatic factors (e.g. temperature, precipitation), agricultural factors (e.g. pesticide use, fertilizer use) and economic factors (e.g. price, production volumes) were collected for the countries of origin of the product concurrent with the period of food safety notification in RASFF. A BN was constructed with 80% of the collected data using a machine-learning algorithm and optimised for each specific hazard category. The performance of the developed BN was determined in terms of accuracy of prediction of the hazard category in the evaluation set comprising 20% of the total data. The accuracy was high (95%) and the following factors contributed most: product category, notifying country, yearly production, number of notification, maximal residue level (MRL) ratio, country of origin, and the annual agricultural budget of a country. The assessment of the impact of interactions within the BN showed a significant interaction between the presence and level of a hazard as reported in RASFF and several drivers of change but at present, no definite conclusions can be drawn regarding the climatic factors and food safety hazards.

Technology is now being developed that is able to handle vast amounts of structured and unstructured data from diverse sources and origins. These technologies are often referred to as big data, and opens new areas of research and applications that will have an increasing impact in all sectors of our society. In this paper we assessed to which extent big data is being applied in the food safety domain and identified several promising trends. In several parts of the world, governments stimulate the publication on internet of all data generated in public funded research projects. This policy opens new opportunities for stakeholders dealing with food safety to address issues which were not possible before. Application of mobile phones as detection devices for food safety and the use of social media as early warning of food safety problems are a few examples of the new developments that are possible due to big data.

Feeding dairy cattle with safe compound feed helps farmers to ensure food safety. However, several ingredients often used in compound feed production can be contaminated with aflatoxin B1 (AFB1), which may result into milk contaminated with aflatoxin M1. Given the number of ingredients and their amounts used in the production of compound feed, it is very costly to check every batch of ingredients for AFB1 contamination. Which is the reason, why a risk-based approach is taken in the latest years. This study aimed to estimate the probability of AFB1 contamination of compound feed for dairy cattle, and to limit this contamination, by optimisation of the compound feed formulation, using a modelling approach. The modelling approach comprised integrating a linear optimisation programming model to a Monte Carlo simulation model. This model was applied to the case of producing compound feed for dairy cattle in the Netherlands, using national monitoring data on AFB1 contamination in feed materials collected in the period 2000-2010. Results from this case study showed the model can be used to produce safe compound feed with the lowest possible probability of AFB1 contamination.

Sampling plans for food safety hazards are aimed to be used to determine whether a lot of food is contaminated (with microbiological or chemical hazards) or not. One of the components of sampling plans is the sampling strategy. The aim of this study was to compare the performance of three different sampling strategies, being simple random sampling (SRS), stratified random sampling (STRS), and systematic sampling (SS), with each other for their probability of detecting a heterogeneously distributed contamination in a lot of herbs or spices (i.e., a dry food product). To this end, a simulation model was developed, and applied to different scenarios for contamination level and numbers of samples collected. In addition, as a case study, the sampling plan of a company processing herbs and spices was evaluated using the simulation model. Results showed that the effectiveness of the sampling plan is influenced by the sampling strategy. With expected low contamination levels the SS strategy performs better than the two other strategies. At higher expected contaminated levels, the STRS strategy is preferred.

Production of sufficient, safe and nutritious food is a global challenge faced by the actors operating in the food production chain. The performance of food-producing systems from farm to fork is directly and indirectly influenced by major changes in, for example, climate, demographics, and the economy. Many of these major trends will also drive the development of food safety risks and thus will have an effect on human health, local societies and economies. It is advocated that a holistic or system approach taking into account the influence of multiple "drivers" on food safety is followed to predict the increased likelihood of occurrence of safety incidents so as to be better prepared to prevent, mitigate and manage associated risks. The value of using a Bayesian Network (BN) modelling approach for this purpose is demonstrated in this paper using food fraud as an example. Possible links between food fraud cases retrieved from the RASFF (EU) and EMA (USA) databases and features of these cases provided by both the records themselves and additional data obtained from other sources are demonstrated. The BN model was developed from 1393 food fraud cases and 15 different data sources. With this model applied to these collected data on food fraud cases, the product categories that thus showed the highest probabilities of being fraudulent were "fish and seafood" (20.6%), "meat" (13.4%) and "fruits and vegetables" (10.4%). Features of the country of origin appeared to be important factors in identifying the possible hazards associated with a product. The model had a predictive accuracy of 91.5% for the fraud type and demonstrates how expert knowledge and data can be combined within a model to assist risk managers to better understand the factors and their interrelationships.

Food fraud is receiving considerable attention with the growing body of literature that recognises its importance. No system exists that collects media reports on food fraud. In this study, we used the infrastructure provided by the European Media Monitor (EMM), in particular it's MedISys portal for this purpose. We developed a food fraud tool (MedISys-FF) that collects, processes and presents food fraud reports published world-wide in the media. MedISys-FF is updated every 10 min 24/7. Food fraud reports were collected with MedISys-FF for 16 months (September 2014 to December 2015) and benchmarked against food fraud reports published in Rapid Alert for Food and feed (RASFF), Economically Motivated Adulteration Database (EMA) and HorizonScan. The results showed that MedISys-FF collects food fraud publications with high relevance >75% and the top 4 most reported fraudulent commodities in the media were i) meat, ii) seafood, iii) milk and iv) alcohol. These top stories align with those found in RASFF and EMA but differences in frequency are apparent. Analysis of the collected articles can help understanding food fraud issues in the origin countries and can facilitate the development of control measures and to detect food fraud in the food supply chain.

While control banding has been identified as a suitable framework for the evaluation and the determination of potential human health risks associated with exposure to nanomaterials (NMs), the approach currently lacks any implementation that enjoys widespread support. Large inconsistencies in characterisation data, toxicological measurements and exposure scenarios make it difficult to map and compare the risk associated with NMs based on physicochemical data, concentration and exposure route. Here we demonstrate the use of Bayesian networks as a reliable tool for NM risk estimation. This tool is tractable, accessible and scalable. Most importantly, it captures a broad span of data types, from complete, high quality data sets through to data sets with missing data and/or values with a relatively high spread of probability distribution. The tool is able to learn iteratively in order to further refine forecasts as the quality of data available improves. We demonstrate how this risk measurement approach works on NMs with varying degrees of risk potential, namely, carbon nanotubes, silver and titanium dioxide. The results afford even non-experts an accurate picture of the occupational risk probabilities associated with these NMs and, in doing so, demonstrated how NM risk can be evaluated into a tractable, quantitative risk comparator.