Prediction of Birth Weight from Body Measurements with the CART Algorithm in Morkaraman Lambs

Cem Tırınk; Ramazan Tosun; Metin Saftan; Emrah Kaya; Ali İhsan Atalay

Cem Tırınk Igdir University, Faculty of Agriculture, Department of Animal Science, Biometry and Genetic Unit, TR76000, Igdir, Turkey
Ramazan Tosun Igdir University, Faculty of Agriculture, Department of Animal Science, TR76000, Igdir, Turkey https://orcid.org/0000-0002-8209-6362
Metin Saftan Igdir University, Faculty of Agriculture, Department of Animal Science, TR76000, Igdir, Turkey
Emrah Kaya Igdir University, Faculty of Agriculture, Department of Animal Science, TR76000, Igdir, Turkey https://orcid.org/0000-0001-7337-0406
Ali İhsan Atalay Igdir University, Faculty of Agriculture, Department of Animal Science, TR76000, Igdir, Turkey https://orcid.org/0000-0002-7379-9082

Abstract

Morkaraman sheep breed is a major breed reared in Eastern Anatolia of Turkey. This study was organized to predict birth weight from biometrical measurements such as withers height (WH), rump height (RH), chest circumference (CC) and body length (BL) taken from 44 Morkaraman lambs. For birth weight prediction, the CART data mining algorithm was performed for ten crossvalidation procedures. Also, the optimum CART tree was achieved with 4 terminal nodes for the lowest RMSE value. The Pearson’s correlation coefficient between the real and predicted birth weight value was determined as 0.935. As a result of the CART algorithm we used, which showed that Morkaraman lambs with CC > 41 cm, male and RH > 41 cm had the heaviest birth weight,
with an average of 5.4 kg. To evaluate the performances of the CART model, the goodness of fit criteria such as RMSE, rRMSE, SDratio, AIC, MAD, RAE, MAPE, R2 and R2 adjwere used. As a result, the model obtained by CART data mining algorithms can be recommended for the estimation of birth weight in lambs, since the determination coefficient is high. In this way, the characterization of the breed will be facilitated, and an important step can be taken for herd management. In conclusion, it will be easier to select a population consisting of traits superior to some biometrical measurements with the CART model.