Comparing the Predictive Ability of Machine Learning Methods in Predicting the Live Body Weight of Beetal Goats of Pakistan

ABSTRACT This study aimed to compare four popular machine learning algorithms to model and predict the body weight through various body measurements of small ruminants. The regression tree, random forests, support vector machine and gradient boosting machine methods have been used to predict the live body weight (target variable) of the animal. The predictors (independent variables) included in the current study were sex of the animal, neck length, diagonal body length, head girth above eyes,shank circumference, belly sprung, and rump height measurements of beetal goats (1 - 7 years in age) of Pakistan. In order to test the performance of candidate methods, various evaluation measures such as the mean absolute error, root mean squared error, mean absolute percentage error, coefficient of determination and the correlation between the actual and predicted body weights were calculated. A 10-fold cross validation was used on the training dataset for tuning the hyperparameters of the models whereas a separate testing dataset was used for evaluation of the predictive performance of machine learning methods. The predictive performance of random forests and gradient boosting methods were found to provide better results than other competing methods in accurately predicting the live body weight of the beetal goats.