Prediction of live body weight based on body measurements in Thalli sheep under tropical conditions of Pakistan using CART and MARS

Thalli sheep is a significant breed reared under tropical region of Punjab province of Pakistan. The present study was conducted to predict live body weight (LBW) by means of from some body measurements, i.e., chest girth (CG), belly girth (BG), rump height (RH), withers height (WH), neck girth (NG), and body length (BL) taken from 155 Thalli indigenous sheep of Pakistan. Age factor is determined to be a significant source of variation for BL, BG, CG, BG, WH, and NG ( p < 0.05). LBW is correlated significantly with BL (0.850), CG (0.825), BG (0.849), RH (0.579), WH (0.547), and NG (0.7760), respectively ( p < 0.01). For LBW prediction, CART and MARS data mining algorithms were comparatively used based on ten cross-validation method. Among 185 candidate MARS models with 1–5 degrees of interaction and 2–38 terms, the MARS model with 7 terms and no interaction effect in R software was the best model for LBW prediction on the basis of the smallest cross-validated RMSE value. Also, the optimal CART tree structure was obtained with 9 terminal nodes for the smallest cross-validated RMSE value. MARS algorithm outperformed CART in LBW prediction and explained 90.3 (%) of variability in LBW of Thalli sheep. Results of the optimal CART structure reflected that Thalli sheep with BL > 75 cm, RH > 83 cm, and NG > 55 cm has the heaviest LBW of 72 kg. The optimal MARS model displays that the heaviest LBW can be produced by Thalli sheep with BL > 71.12 cm, BG > 106.68 cm, WH > 76.2 cm, NG > 50.8 cm in 5 th age group. In conclusion, it coud be recommended that MARS predictive modeling may enable animal breeders to obtain elite Thalli sheep population and to detect body measurement positively influencing LBW as indirect selection criteria for not only describing breed characterization and developing flock management standards, but also ensuring sustainable meat production and rural development in Pakistan.