Use of linear modeling, multivariate adaptive regression splines and decision trees in body weight prediction in goats

Abstract

Use of robust regression algorithms for better prediction of body weight (BW) is receiving increased attention. The present study therefore aimed at predicting BW from chest circumference, breed and sex of a total of 1,012 goats. The animals comprised 332 matured West African Dwarf (WAD) (197 bucks and 135 does), 374 Red Sokoto (RS) (216 bucks and 158 does) and 306 Sahel (SH) (172 bucks and 134 does) randomly selected in Nasarawa State, north central Nigeria. BW prediction was made using automatic linear modeling (ALM), multivariate adaptive regression splines (MARS), classification and regression tree (CART), chi-square automatic interaction detection (CHAID) and exhaustive CHAID. The predictive ability of each statistical approach was measured using goodness of fit criteria i.e. Pearson?s correlation coefficient (r), Coefficient of determination (R2), Adjusted coefficient of determination (Adj. R2), Root-mean-square error (RMSE), Mean absolute percentage error (MAPE), Mean absolute deviation (MAD), Global relative approximation error (RAE), Standard deviation ratio (SD ratio), Akaike?s information criterion (AIC) and Akaike?s information criterion corrected (AICc). Male RS and SH goats had significantly (P<0.05) higher BW and CC compared to their female counterparts while in WAD, male goats had significantly (P<0.05) higher CC (57.88?0.51 vs. 55.45?0.55). CC was determined to be the trait of paramount importance in BW prediction, as expected. Among the five models, MARS algorithm gave the best fit in BW prediction with r, R2, Adj. R2, SDratio, RMSE, RAE, MAPE, MAD, AIC and AICc values of 0.966, 0.933, 0.932, 0.26, 1.078, 0.045, 3.245, 0.743, 186.0 and 187.0, respectively. The present information may guide the choice of model which may be exploited in the selection and genetic improvement of animals including feed and health management and marketing purposes, and especially in the identification of the studied breed?s standards.