The administration of additional painkillers during/after disbudding was also more common on organic farms: all but three organic farms provided this additional care option, whereas in half of the conventional farms no analgesic was administered for disbudding . These differences were probably due to the legal standards and the animal friendly management measures prevalent in organic farming systems.Companies in the livestock sector are increasingly investigating the potential of data-driven decision making for improving production . In the pork sector, the growing demand for pork meat and the outbreak of diseases that threatens pig farming have forced pig farmers to implement precision farming practices. The rapid implementation of precision farming following the outbreak of African swine fever and the subsequent shortage of pork presents an illustrative example of the value of data-driven decision making . Data-driven decision making at pig farms means that decisions that are made will depend on predictions made using the information gathered at the farm and across the supply chain. To perform successful predictions and help in decision making, data analytics and machine learning techniques can be used. Recently, ML models are being used to predict various variables of interest to decision making, such as sales and feed performance . Data gathering and analytics are thus becoming crucial for gaining a competitive advantage in the agricultural sector in general, and in the pork sector in particular. The availability and accessibility of structured data are crucial for the performance of ML models . It is therefore important to investigate the entire pipeline of data analytics in order to apply improved data analytics in practice.There have been only a few systematic reviews of the literature with respect to data analytics in the pork sector. Koketsu and Iida reviewed the literature on the data analysis used for the performance of sows and predictors in breeding herds.
The review defined four components of sow lifetime performance, namely lifetime efficiency, sow longevity, fertility,greenhouse benches and prolificacy. Further, they proposed two different lifetime performance trees, one for piglets weaned and one for pigs born alive, which have relationships with the performance components. They also described predictors for high lifetime performance. Nguyen Thi Thuy et al. did a review on the pork value chain in Vietnam. They concluded that the presence of middlemen and traders is important for facilitating the nationwide distribution of pigs and pork, but they also observed that, in general, no official contracts are used in transactions in the current supply chain, and as a result, it is hard to trace the origin of pork in Vietnam. Newer pork value chains are developing in Vietnam but currently represent only a fraction of the whole pork supply chain. The purpose of this study is to perform a Systematic Literature Review and find out what research has been done on the application of data analytics, data-driven management, and data-driven decision support in the pork sector and identify the associated data used. Animal-based products have a large and increasing contribution to the human food supply . The increasing demand is driven by the increase of both the world population and the wealth of households. Optimizing animal-based products requires optimizing the feed intake of animals, reducing the risk of diseases, and improving the animals’ living environment and their general welfare state. Optimizing animal-based production systems requires, therefore, coordination among farmers, feed companies, and other actors. The ultimate aim of optimized animal-based production is to attend to the precise needs of the individual animal, which is called precision livestock farming . The transition to PLF is achieved when extensive data on animal growth, production outputs, disease development, animal behaviour, and living environment are gathered and analysed for use in data-driven decision making . PLF is part of the general trend of precision farming which is practiced in diverse disciplines of agriculture, such as arable farming . Machine learning , which includes the application of deep learning and neural networks, is a branch of artificial intelligence . AI is a key element of the fourth industrial revolution, and in agriculture this revolution is referred to as Agriculture 4.0. Different kinds of agricultural technology are being used for many years, but the introduction of IoT, cloud computing, robotics, and AI could change the way of farming significantly.
A variety of research has been done on the use of ML in the agricultural sector, including in the livestock sector. For example, cows’ body weight was predicted using a support vector machine classification model, and several ML algorithms have been used in disease detection . Table 1, derived from Neethirajan , shows which diseases have been detected using ML. Optimizing feed efficiency is crucial because feed is the biggest cost factor in animal production . Table 2 shows what algorithms have been used and how data was collected for optimizing feed consumption. Besides feed efficiency and disease detection, ML has been used in the livestock sector for behaviour tracking, health monitoring, and the management of the supply chain as shown in Table 3. When looking specifically at the research on pigs, most attention has been given to the growth of pigs. Much research is done to find patterns in behaviour, the environment, and weight gains. For example, patterns have been found that the more pigs contract Pneumonic Pasteurellosis, a disease that occurs throughout the world in pig farming, the less will be their daily weight gain . Researchers at Wageningen University are currently working on the ‘Feed-a-Gene’ study, focusing on the improvement of feed conversion. Their latest findings showed that when pigs get adjusted nutrients per pig, the impact on the environment will be lower. An alternative feed system was developed and pigs were fed based on their individual characteristics . Besides the livestock sector, ML has multiple other applications in the agricultural sector. ML has been used in arable farming for a variety of prediction problems. For example, crop yield prediction is done using ML and data gathered over the soil conditions, temperature, and rainfall . Precision farming is applied successfully in the arable farming sector .Information systems are used in the meat supply chain sector for several reasons. Farmers use farm management information systems to manage their herds. Bigger companies, such as feed producers, slaughterhouses, and meat processors use enterprise systems. Enterprise systems are often complex systems and are thus modular and configurable. The aspects of information systems relevant for this research are the use of data for decision support and data integration across the supply chain. A supply chain of a product is a set of coordinated entities that are involved in the realization of the product . When looking at the supply chain from the perspective of one of the actors , upstream and downstream flows can be defined. The upstream supply chain is focused on supplying the needs for production, and the downstream chain is focused on the flow of products to its end consumers. With the implementation of more sensors and data sources, more possibilities arise for using ML in information systems.
Manufacturing companies are often focused on static explanatory models, but there are many opportunities in predictive modelling where ML can be used and improve the decision making processes . For example, ML is used to predict limb conditions of pigs, using data collected at farm . Liang et al. predicted African swine fever outbreaks, using ASF outbreak data and meteorological data. Xu et al. proposed a pork traceability framework, based on IoT monitoring and data mining techniques. A prerequisite for applying improved data analytics, such as ML, is the availability of large and high quality dataset . Businesses gather data for various purposes but not all of it is good enough for use in ML. The data often has inaccuracies and inconsistencies, also known as noise . When data is collected from various databases in the supply chain, the data from different sources have to conform to a common standard. Otherwise, combining the data from the different sources will be difficult or even impossible. When certain data are gathered at one place but the related data are not gathered at the other places , the desired objective of data analytics may not be achieved. In the last decades, data were mostly used for the management of farm tasks, and to a limited extent for further analytics . The variables that are routinely measured were batch data on growth , feed intake, and mortality. Feed efficiency and performance were determined using these data. Environmental data and data from slaughterhouses were not widely available or accessible and thus were not used widely. Feeding costs have a significant share in the total production costs for animal production . When the feeding process could be optimized, the financial impact could be significant. When the collection and analysis of useful information is made available to farmers and integrated within their farm information management systems,plant benches productivity could become higher.Technologies that identify the individual animal and collected data accordingly can help farmers to monitor individual animals and improve the welfare of the animals. When data is visualized in meaningful and easy-to-understand ways, social acceptance of the technologies could become higher.Since feed has the largest share in the variable production costs in livestock farming, optimization of the feed use could result in more sustainable and controlled production.
Machine learning is used in diverse business processes in a variety of companies. For example, data mining with ML algorithms is done in finance, telecom, marketing, and web analytics firms. The application of ML has produced useful results when used for the prediction of various variables but there are also many aspects that can be improved. There is a trade-off between the reward of using an ML application and the investment that comes with it, while successful prediction cannot always be guaranteed. In the finance sector, for instance, ML is more widely used to reduce uncertainty when making decisions but the accuracy of the predictions depends on a number of factors and the results should be used cautiously . After finishing the planning phase, the review was conducted by searching the databases using search strings that are adjusted according to the options the databases offer. When the search in a particular database returned more than 50 publications, only the first 50 most relevant publications were selected. The publication details were saved, including title, author names, type of publication, and publication year, merged, duplicates removed, and filtered using exclusion criteria. After the selection of relevant publications, the selected publications have been analysed, also known as primary studies. After conducting the review, the last step in the process was executed. The results of the review have extensively been reported using tables and used to answer the research questions. The search is performed in five well-known scientific databases: Science Direct, Scopus, Web of Science, Springer Link, and Wiley. The search was done by choosing a broad search string and incrementally narrowing down the search. We started by searching for “Data use in agriculture” in Science Direct, which returned 23.217 results which shows this search query is too broad. Therefore, we limited the search string to make it focus on the livestock sector and the specific issue that the research questions addressed. However, when a search query was narrowed further no results were returned. Finally, we selected a broad enough search string and selected the 50 most relevant publications using the relevance order offered by the database. This was done to exclude less relevant publications; also, publications that focus on the end-of-chain consumer side were excluded since most of them focus on the storage of meat and sales related publications. When using a feasible search query, the abstract of the study was read to see if the source was of relevance. If this was the case, the complete study was read and the study was analysed based on the exclusion criteria. Finally, the data that is required for answering the research questions were extracted and analysed. The large result of the search queries indicates that we chose our search string in such a way that we obtained a broad perspective on the research done in the pork chain. This large initial result allowed us to browse the results and get an impression of the research done.