Improving herd-based cattle weight gain forecasting using non-time-series data within a hybrid machine learning and deep learning framework
Smart Agricultural Technology · 2026-12 · 15 · 102448
DOI: 10.1016/j.atech.2026.102448 ↗Kurz gefasst
Fachartikel in der Zeitschrift Smart Agricultural Technology, 2026-12. Verfasst von Muhammad Riaz Hasib Hossain und 4 weiteren Autor:innen. Die Arbeit nennt 41 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Charles Sturt University CSU, Food Agility CRC.
Abstract (Originalsprache)
Reliable forecasting of herd-based cattle weight gain supports feeding management, pasture utilisation, and marketing decisions in grazing systems. Conventional forecasting approaches rely on time-series models that assume regular observation intervals. Step-On-Off (SOO) weighing datasets often contain irregular sampling intervals and missing observations, which pose challenges for conventional time-series forecasting methods. This study evaluates a hybrid Gradient Boosting-Neural Network (GB-NN) framework for weekly cattle weight forecasting using transformed non-time-series predictors. Data collected between 2022 and 2024 at the Charles Sturt University Global Digital Farm were integrated, including Optiweigh live-weight measurements, animal demographic attributes, and lagged rainfall and temperature variables. Forecasting performance was benchmarked against established time-series approaches, including Autoregressive Integrated Moving Average, Prophet, and Long Short-Term Memory models, using identical animal-level holdout validation. The hybrid framework achieved 97.05% mean prediction accuracy across one- to twelve-week forecasting horizons under independent testing conditions. Predictive performance consistently exceeded all baseline models. Animal age, previous live-weight, rainfall, and temperature emerged as dominant predictors of herd-level growth dynamics. Lagged environmental variables and prior weight information enabled structured predictors to represent short-term temporal dynamics in cattle growth. Results demonstrate that biologically informed tabular predictors combined with residual learning enable accurate multi-week forecasting without explicit sequential inputs. The proposed framework provides a practical foundation for decision-support systems supporting feed allocation, pasture management, anomaly detection, and livestock marketing in commercial grazing enterprises.
Förderung
- Charles Sturt University CSU
- Food Agility CRC
Zitieren
APA
Hossain, M. R. H., Islam, R., McGrath, S. R., Islam, M. Z., & Lamb, D. (2026). Improving herd-based cattle weight gain forecasting using non-time-series data within a hybrid machine learning and deep learning framework. Smart Agricultural Technology, 15, 102448. https://doi.org/10.1016/j.atech.2026.102448
Vancouver
Hossain MRH, Islam R, McGrath SR, Islam MZ, Lamb D. Improving herd-based cattle weight gain forecasting using non-time-series data within a hybrid machine learning and deep learning framework. Smart Agricultural Technology. 2026;15:102448. doi:10.1016/j.atech.2026.102448
BibTeX
@article{Hossain2026,
author = {Hossain, Muhammad Riaz Hasib and Islam, Rafiqul and McGrath, Shawn R. and Islam, Md Zahidul and Lamb, David},
title = {Improving herd-based cattle weight gain forecasting using non-time-series data within a hybrid machine learning and deep learning framework},
journal = {Smart Agricultural Technology},
year = {2026},
volume = {15},
pages = {102448},
doi = {10.1016/j.atech.2026.102448}
}
EUVIMED