The Cobb R&D team collaborated with researchers at the University of Georgia and the University of Madison-Wisconsin to determine if feeding behavior could be used to predict broiler chicken health. Animals, chickens included, will change their eating patterns, social interactions, and general activity when they become ill or injured. Using this knowledge, Cobb R&D and collaborators designed an experiment to track birds with radio-frequency identification (RFID) transponders. The birds were fitted with RFID tags on their wings and the feeding stations had antennas. The feeding stations, designed by Cobb’s engineering team, allowed only 1 bird to enter at a time. In this way, the researchers were able to track and discern individual bird metrics.
The experiments were conducted over a 5-year period and followed more than 95,000 broiler chickens, ultimately collecting data on nearly 100,000,000 feeder visits. The data was summarized into different feeding traits (feed intake, number of visits, time spent at the feeder, time between feeder visits, and number of feeders visited). The feeders were equipped with scales so feeding amount (g/meal) and feeding rate (g/hour) were also measured.
Schematic of the experimental design to collect feeding behavior raw data. During the feeder visit the transponder on the bird’s wing is activated by the antenna and transmits data, decoded by the reader, and sent to the local server. The processed information was stored in a cloud database.
They used the data to train and test 5 different machine learning models, each with a different classification system to try and uncover subtle patterns indicative of illness. Two of the algorithms (GBM and SVM) provided the best fit, with mortality predictions being made 1 to 3 days before the actual event.
With this type of system in place at the farm, early intervention to treat ill or injured birds could improve animal welfare outcomes and reduce economic losses. Moreover, automated monitoring favors biosecurity by reducing the need for in-house flock inspections. Last, the system can collect and store data allowing future management decisions to be well informed and precise.
Article citation:
Alves, Anderson AC, Arthur FA Fernandes, Vivian Breen, Rachel Hawken, and Guilherme JM Rosa. “Monitoring mortality events in floor-raised broilers using machine learning algorithms trained with feeding behavior time-series data.” Computers and Electronics in Agriculture 224 (2024): 109124.
