Comparison of some non-linear functions to describe the growth for Linda geese with CART and XGBoost algorithms

Abstract

The aim of this study was to determine the best non-linear function describing the growth of the Linda goose breed. To achieve this aim, five non-linear functions, such as exponential, logistic, von Bertalanffy, Brody and Gompertz, were employed to define the live weight-age relationship for male and female Linda geese. In the study, 2 397 body weight-age records from 75 females and 66 males collected from three days to 17 weeks of age were evaluated using the “easynls” and “nlstools” packages for growth modelling of the Linda goose in R software. Each model was analysed in the live weight records of all the geese separately for males and females. To measure the predictive quality of the growth functions used individually here, model goodness of fit criteria, such as the coefficient of determination (R²), adjusted coefficient of determination (R²_adj), root mean square error (RMSE), Akaike’s information criterion (AIC) and Bayesian information criterion (BIC) were implemented. Among the evaluated non-linear functions, von Bertalanffy model gave the best fit of describing the growth curve of female and male Linda geese. Based on the “rpart”, “rpart.plot”, and “caret” R packages, the CART and XGBoost algorithms were specified in the prediction of live weight of Linda geese at 17 weeks of age from the growth parameters of the von Bertalanffy model and the sex factor. XGBoost produced better results in superiority compared with the CART algorithm. In conclusion, it could be suggested that the von Bertalanffy model might help geese breeders to determine the appropriate slaughtering time, feeding regimes, and overcome flock management problems. The results of the XGBoost algorithm might present a good reference for breeders to establish breed standards and selection strategies of Linda geese in the growth parameters for breeding purposes.