Comparative analysis of machine learning algorithms for predicting live weight of Hereford cows

Abstract

Body weight prediction of livestock helps us to control the health of animals, efficiently conduct genetic selection, and estimate the optimal slaughter time. Precise and expensive industrial scales are used to measure live weight on large farms. A more affordable alternative is weight estimation by indirect methods, based on morphometric measurements of livestock, followed by the use of regression equations relating such measurements to body weight. Manual measurements on animals with a tape measure require trained workers and are stressful for both the worker and animal. Nowadays, machine learning technologies are being used to accurately predict body weight. This paper provides a comparative analysis of various machine learning methods for estimating the live weight of Hereford cows in terms of the coefficient of determination, root mean squared error, mean absolute error, and mean absolute percentage error. We show that machine learning algorithms perform better than common linear regression algorithms. Specifically, the ExtraTreesRegressor algorithm yields the highest prediction quality of the live weight of Hereford cows in terms of $R^2$ among the tested machine learning algorithms. Potential applicability of these methods in the livestock industry is also discussed.