Prediction of body weight in indigenous sheep using random forest regression, support vector regression, and classification and regression trees algorithms

This study aimed to develop predictive models for estimating the body weight (BW) of indigenous sheep in Ethiopia using morphometric traits, comparing the performance of random forest regression (RFR), support vector regression (SVR), and classification and regression trees (CART) algorithms. Data were collected from 306 mature sheep (249 ewes and 57 rams) in Tahtay Maichew district of Tigray region, including BW and 16 linear body measurements (LBMs). Descriptive statistics indicated low to moderate variability in BW and key LBMs, with heart girth (HG) and body length (BL) showing the strongest correlations with BW (<i>r</i> = 0.61 and 0.46, respectively). Among the models evaluated, RFR demonstrated superior predictive accuracy, achieving the highest <i>R²</i> values (0.809 training, 0.477 validation) and the lowest root mean square error (RMSE: 1.650 training, 2.825 validation). SVR performed well in training but had lower generalizability in validation, while CART showed the weakest performance overall. Variable importance analysis identified HG as the most influential predictor across all models, contributing 40.90 % in RFR, 16.30 % in SVR, and 68.76 % in CART, often followed by BL and height at withers (HAW). The findings highlight the potential of RFR as a robust tool for BW prediction in resource-limited settings where weighing scales are unavailable. The study provides practical insights for smallholder farmers and breeding programs, enabling improved genetic selection and management practices based on easily measurable morphometric traits, particularly HG. Future research should validate these models with larger, more diverse datasets to enhance their applicability across different agroecological zones and sheep populations.