Machine-learning-based quantitative body condition scoring in Qinchuan beef cattle using conventional body measurements.

Body condition score (BCS) is a cornerstone indicator for beef cattle reproduction and nutritional management, but conventional visual scoring is subjective and lacks quantitative standards. This study aimed to develop a low-cost, objective BCS evaluation system for Qinchuan beef cattle using routine body measurements and machine-learning models. A total of 321 qinchuan breeding cows aged 24-36 months were scored for BCS (4-8) by two trained evaluators following the Chinese national standard, and seven conventional traits (withers height, body length, hip height, heart girth, abdominal girth, rump width, and body weight) were recorded. One-way ANOVA and Pearson correlation analyses were used to test the biological consistency of subjective scores and to identify informative traits. Six composite body condition indices were constructed to standardize frame size, and seven machine-learning algorithms were compared under nested cross-validation using quadratic weighted kappa (QWK) as the primary metric. All measured traits increased significantly across BCS classes, and heart girth, abdominal girth, and body weight showed the strongest correlations with BCS ([Formula: see text], [Formula: see text]). A composite morphometric index (CMI) provided the best trade-off between correlation and classification performance. The best model, k-nearest neighbors using six key traits, achieved a QWK of 0.89 and 95.5% prediction accuracy within ±1 BCS unit. This study demonstrates that conventional body measurements, combined with composite indices and machine-learning models, can provide a practical, scalable tool for objective BCS evaluation in Qinchuan beef cattle and support precision livestock farming.