Imagine a future where the health status and body weight of every pig on the farm could be accessed individually on a mobile phone app; a future where sensor networks would continually monitor pig behaviour, feed consumption and barn climates in real time, and could alert producers when these variables deviated from the normal range. This future is rapidly becoming a reality with the arrival of precision livestock farming and big data technology, and this holds many exciting opportunities for the pork industry.

In precision livestock farming, the kind of vast, complex data sets that can be mined and analysed by big data technology can be generated by sensor networks in the animal houses. Constant monitoring provides granular, detailed information that can be processed by a computer to be used by producers to optimise the management of their herds. For example, using technology to categorise a pig as lame based on the proportion of time it spent lying down is a much more useful measure than simply providing the farmer with the total amount of time that the pig spends on the floor.

The ability to detect patterns in big data has been significantly strengthened by the advent of machine learning. Machine learning involves computer programs and algorithms that can adapt over time to recognise what is normal in a data set by being exposed to large amounts of data. The computers can then identify any data that is abnormal, and report which conditions led to this deviation. The more data that they are exposed to, the more accurate their predictions become. This results in the ability to predict events such as disease outbreaks, or notify producers when there is a problem in the barn that needs their intervention.

Machine learning was recently used to identify previously unknown patterns of disease transmission on US pork farms (Scientific Reports (2019) 9:457). It was already known that moving animals between farms, such as transferring piglets to grower farms, was associated with an increased risk of porcine epidemic diarrhoea virus (PEDv), but the transmission of the disease between neighbouring farms in the absence of animal movement was not understood. The authors suspected that climatic variation was involved. To investigate this theory, the transport data of over 15 million pigs originating from 332 sow farms was combined with data on the concurrent weather patterns, the physical features of the farms and nearby landscapes and pig population density to identify patterns associated with disease outbreaks. Machine-learning algorithms confirmed that animal transport was a high risk factor for PEDv, but also identified aspects of the weather such as season and windspeed that were associated with PEDv outbreaks. These results are consistent with the airborne transmission of this disease, a finding that would not have been possible using traditional statistical methods.

Back on our farm of the future, the capacity for computers to remotely detect changes in pig behaviour are mind boggling. For example, cameras located above the pens can now accurately track the behaviour of groups of pigs to detect behaviours such as huddling, feeding, fighting and tail biting. Currently these behavioural measurements must be made at the pen level, as the tendency for pigs to lie close together precludes individual tracking. For lactating sows that are housed individually, video imaging can be used to monitor nursing behaviour and piglet-crushing events. Image analysis can also be used to estimate grower body weight to within 1kg, allowing age at market weight to be predicted more accurately. Thermal imaging can be used to detect changes in body heat indicative of illness, and audio sensors can detect heat stress, pain and respiratory disease based on the vocalisations and coughing emitted by a group of pigs. Cameras can even identify individual pigs using facial recognition!

With access to the appropriate computing power, this wealth of digital information can be used to rapidly detect health and welfare problems in the herd. For example, ear tags containing accelerometers can track pig movement and can automatically detect lameness 14 days prior to signs of moderate lameness developing. As it is often difficult to manually detect lameness in large groups before it becomes moderate to severe, this technology has the potential to decrease the time taken for lame pigs to receive treatment and reduce the duration of their suffering. Individual monitoring of production data, such as feed-conversion efficiencies and growth rates, may also help with tailored management decisions and improved profitability.

The rise of precision livestock farming is the inevitable next step in the livestock industries. Despite the obvious benefits for producers, there is a range of physical and technical difficulties that must be overcome before this technology can become widespread. For example, the sensors used must be robust enough to withstand life inside an animal house, and the initial cost of installing these may present a barrier to some. Producers will also need access to computer systems that are capable of processing big data into useable outputs, and the poor internet connectivity in many rural areas may present further barriers for solutions that rely on remote technology. However, many of these issues will be addressed as this technology develops in the future.

For computers to develop the algorithms needed to process big data into useful outputs, the machines must first be “trained” using large data sets, and the results validated by scientists in terms of their biological relevance for pig health and welfare. Access to big data for machine learning requires collaboration and data sharing within the industry, and this presents both advantages and disadvantages for producers. Some forms of big data are publicly available, such as animal disease incidence data or weather data. Other data is only available privately, and this includes production data and other data that businesses may wish to be kept private in a competitive environment. The future development of externally curated, privacy-protected repositories for the storage of private data may help to alleviate these concerns while allowing access to this data for machine learning.

In conclusion, the use of big data and machine learning to monitor the physiology and behaviour of pigs in real time presents many opportunities to improve their health and welfare. Problems in the animal house can be identified and rectified more rapidly, and potential problems can be more easily predicted. The immediate response of the pigs to changes in their environment, such as the provision of enrichment or the removal of a stressor, can be measured in real time and the husbandry of these animals adjusted accordingly. Precision livestock farming combined with machine-learning technology can help producers to provide individualised care to their animals while simultaneously managing them in large numbers.