Comparison of MARS, CART, and Linear Regression Models for Prediction of Body Weight of Non-descript Indigenous Goats in Lepelle-Nkumbi Local Municipality, South Africa

Abstract

In Lepelle-Nkumbi Local Municipality of South Africa, 200 none-descript indigenous goats ranging in age from one to five years were the subjects of a study that compared the live body weight predictions made by stepwise linear regression, Classification Regression Tree (CART), and Multivariate Adaptive Splines (MARS) models. Several bodily measurements, such as canonical circumference (CC), sternum height (SH), body length (BL), ear length (EL), head length (HL), head width (HW), rump length (RL), rump height (RH), and rump width (RW). The evaluation criteria included the root mean square error (RMSE), coefficient of determination (R2), to decide which model was the best. According to the results, CART outperformed the others, obtaining the lowest RMSE (3.65) and the greatest R2 (0.80). The stepwise regression model outperformed data mining algorithms in male goats. According to the study, CART is a useful statistical technique for defining requirements for producing indigenous goats that are not very special. In addition, when predicting live body weight from body measuring features, the stepwise regression model should be considered.