AI-based framework to predict animal and pen feed intake in feedlot beef cattle

Researchers from the University of Idaho and collaborators have developed an AI-based framework to predict feed intake in feedlot beef cattle, published on arXiv (2511.17663). The system uses XGBoost, a gradient-boosted decision tree model, trained on over 16.5 million samples from 19 experiments conducted between 2013 and 2024 at the Nancy M. Cummings Research Extension & Education Center in Carmen, ID. The model integrates environmental data from AgriMet weather stations with electronic feeding system records to predict intake at both individual animal and pen levels, achieving root mean square errors of 1.38 kg/day for individual animals and 0.14 kg/(day-animal) for pen-level aggregation.

The framework introduces two novel environmental indices: InComfort-Index, based solely on weather variables, which effectively predicts thermal comfort but has limited feed intake accuracy; and EASI-Index, a hybrid index that combines environmental conditions with feeding behavior data, which excels at predicting feed intake but is less suited for thermal comfort assessments. This dual-index approach allows the system to adapt predictions based on specific management goals. The work fills a gap in precision livestock farming by providing a methodology that fully leverages longitudinal big data from electronic feeding systems, enabling autonomous, climate-adaptive management. Potential applications include reducing feed waste, optimizing resource allocation, and improving sustainability in beef production through real-time, data-driven decisions.