Ensemble machine learning for accessible livestock weight prediction in West African cattle from resource-constrained regions

This study presents a comparative evaluation of machine learning models for predicting live body weight in West African cattle using morphometric measurements, with emphasis on applicability in resource constrained environments. Data from 172 cattle representing four breeds were analyzed using 16 morphometric traits. Correlation analysis identified strong associations between body weight and hip width r = 0.95, thoracic girth r = 0.93, and occipito ischial length r = 0.92, while also revealing multicollinearity among several thoracic traits. Fifteen machine learning models were assessed using R², RMSE, MAE, and Lin’s Concordance Correlation Coefficient. The Voting Regressor provided the most favorable balance between predictive accuracy and generalization, achieving test R² = 0.9891, RMSE = 9.85, MAE = 7.80, and LCCC = 0.994, with stable cross validation performance. Permutation based feature importance identified thoracic girth (PT), hip width (LH), and Shoulder Width (LE) as the most influential predictors. A reduced three trait model maintained strong performance with R² = 0.960 and LCCC = 0.979, substantially simplifying measurement requirements while preserving predictive reliability. These findings demonstrate that accurate and robust body weight estimation can be achieved using a small set of easily measurable traits, providing a practical foundation for deployable decision support tools in low resource livestock production systems.