Prediction of body weight in buffaloes using conventional and multi-view image-based morphometric traits

Accurate body weight estimation is essential for nutritional management, growth monitoring, drug dosage determination, and genetic evaluation in buffalo production systems. However, direct weighing of buffaloes is often labor-intensive and difficult to implement under field conditions. Therefore, this study evaluated the potential of conventional and image-derived morphometric traits for predicting body weight in Khuzestani buffaloes using an Elastic Net (EN) regression framework. A total of 100 female buffaloes were evaluated using manually collected body measurements and image-derived traits extracted from side-view and top-view photographs. EN models were optimized using leave-one-out cross-validation (LOOCV) combined with repeated 10-fold cross-validation for hyperparameter tuning. Both manually measured and image-derived traits showed strong positive associations with body weight. Among manually measured traits, barrel girth (BG), chest girth (CG), and rump girth (RG) showed the highest predictive ability, whereas barrel width (BW) was the most informative image-derived trait. The final EN model based on manually measured traits achieved high prediction accuracy, with an average relative error (ARE) of 3.82%, correlation of 0.985, and R2 of 0.970. Similarly, the image-based EN model showed strong predictive performance, with corresponding values of 4.13%, 0.98, and 0.960, respectively. Overall, the results demonstrate that integrating multi-view image phenotyping with EN regression provides an accurate and non-invasive framework for body weight prediction. This study serves as a proof-of-concept for the development of practical computer vision tools, although further validation under commercial farm conditions is required before routine field application.