Hybrid ML Framework for Cattle Growth Forecasting in Grazing Systems

A recent study published on arXiv (2608.06001) presents a hybrid machine learning approach aimed at forecasting herd-level cattle growth and weight gain within grazing-based production systems. Conducted in southeastern Australia from 2022 to 2024, this research tackles the issue of sporadic livestock observations common in commercial grazing operations. The proposed framework combines weekly live weight data, demographic factors, and delayed environmental indicators into organized forecasting datasets. By aggregating animal-level predictions over time, herd-level trajectories are created. The study assessed four families of hybrid architectures: residual, stacked, cascade, and ensemble-assisted, using ARIMA, LSTM, and GRU models as benchmarks. The cascade architecture (GB to RF to NN) yielded the highest performance, achieving a test R² of 0.889 and an RMSE of 21. This work provides a solid method for enhancing cattle weight forecasting, which can improve management strategies in grazing systems.