Modeling and predicting the growth of indigenous Harnai sheep in Pakistan: non-linear functions and MARS algorithm

Abstract

Five non-linear functions, i.e. Gompertz, Logistic, Negative exponential, Brody and Bertalanffy, and multivariate adaptive regression splines (MARS) data mining algorithm were implemented with the objective to describe the body weight-age relationship of Harnai sheep of Balochistan, Pakistan. The data comprised of 1317 records of body weight from birth to 1 year were provided from Multi-Purpose Sheep Research Station Loralai, Balochistan. Each non-linear function and MARS algorithm were fitted to the data of male and female, single and twin and all lambs. Comparison among different non-linear models was based using the adjusted coefficient of determination ([Formula: see text]), Durbin-Watson statistic (DW), root mean square error (RMSE), Akaike's and Bayesian information criteria (AIC and BIC) and the coefficient of correlation (r) between observed and fitted live body weight. The best fit was provided by the Brody model in terms of the highest [Formula: see text] and r values and lowest RMSE, AIC and BIC values in male and female, single and twin and all lambs followed by Bertalanffy, Gompertz, Negative exponential and Logistic model in order of their goodness. The negative correlation between asymptotic weight and maturing rate inferred that animals with smaller mature weight mature fast. Though males and singles were found heavier at mature weight than females and twins, respectively, they mature more slowly. The results of the study suggested the use of the Brody model to accurately describe the weight-age relationship of Harnai sheep. The present study also showed a very high predictive performance of the MARS data mining algorithm for describing the growth of sheep. In conclusion, MARS algorithm may be a good alternative for breeders aiming at describing the weight-age relationship of Harnai sheep.