Using machine learning to predict on-farm challenges and optimize broiler performance

The broiler industry is very good at collecting data from a wide range of input points including house temperature, humidity, CO2, ammonia, feed and water intake, an array of flock settlement data and now blood biomarkers. But collecting and storing data alone is not very helpful. It is only when the data is analyzed, and the insights applied to make improvements, that the value of the data becomes apparent. Increasingly, advanced machine learning tools are being used to help veterinarians and nutritionists identify and tackle challenges before they impair bird health and performance.

Nutritionists are spending more time and becoming more skilled in data analysis. They are expected to interpret performance data and adjust dietary formulations to improve output as quickly as possible. The technological advancement of machine learning is equipping nutritionists and veterinarians with a powerful tool for data analysis and flock optimization. Without machine learning, the broiler industry would otherwise miss the opportunity to turn their costly data collection efforts into a source of performance improvement.

Machine learning in practice

One example of how machine learning has been valuable in practice for a US broiler integrator is in the use of blood biomarkers to highlight a problem with chloride levels (Figure 2).

Using dsm-firmenich’s Verax™ , a model was created to predict acid-base balance in the birds. Chloride levels are highly regulated in broiler production, so to see chloride levels above the expected, normal level indicates there was an issue, perhaps kidney damage, high salt content in the water or high salt content in the feed. On further discussion with the producers, they reported that the floors of the houses were wet, resulting in burnt footpads. Wet litter can be caused by an imbalance of chloride as well as sodium. When the diet was adjusted to limit chloride intake, the floors dried out and there was a 20% improvement in footpad quality which increased the overall profitability of the integrator.

Machine learning to identify the cause of field rickets

Field rickets could be the result of a nutritional problem or a veterinary problem. Without data, responsibility is sometimes passed from nutritionist to veterinarian with no real action being taken to address the underlying issues. However, when data from farms experiencing field rickets was analyzed using Verax™ , a steady drop in phosphorus was seen over time (Figure 3).

Using the insight gained from the blood biomarker data a feed mill visit revealed an error in batching due to a needed scale adjustment. Once the scales were fixed, the incidence of rickets subsided. Without the insights of the data analysis from Verax™, the increased mortality due to field rickets could have been very costly.

Machine learning and early detection of coccidiosis

By the time clinical symptoms of coccidiosis are seen in the flock, feed conversion and other performance parameters will already be negatively affected. Using a Verax™ model based on blood biomarker changes, a highly accurate prediction could be made for coccidiosis 7-10 days before any clinical signs are apparent. Giving producers the ability to intervene earlier can reduce the economic impact of the disease which has been estimated at more than US$1 billion per year.

Conclusion

The broiler industry needs precision analytics technology in order to transform the cost of data collection into a source of competitive advantage. Models can be used to predict problems, allowing interventions to be applied ahead of clinical signs, therefore minimizing losses.

The focus of precision technology should be on the program level, and not on individual houses or individual birds. Machine learning is not about comparing one house to another, it is about looking at the bigger picture and identifying problems as early as possible. Our view must expand beyond individual research trials and individual farms.

The real value of the prediction tool is realised when considering the following questions:

• What is feed cost and the price of chicken in your market?
• What is the value of 1 point in feed conversion in your operation?
• What is the value of 1% livability in your operation?

Broiler producers will know what each percentage point of FCR is worth to the company. Precision technology can help producers achieve those improvements. Looking ahead, work is also being done to include the microbiome. The microbiome is a very complex system, but research is uncovering some insights which will be combined with other parameters including performance data as well as blood biomarkers, increasing the value of such models in supporting performance improvements.

Machine learning systems like Verax™ can be used to put all the data we compile in the industry to work, thereby improving animal health, welfare, performance and sustainability.