Use of Machine Learning Approaches for Body Weight Prediction in Peruvian Corriedale Sheep

Abstract

The goal of this study was to predict the body weight of Corriedale ewes using machine learning (ML) algorithms. Fourteen body measurements (BM) and six different machine learning models were used. Body weight (BW) and BM: wither height (WH), rump height (RH), thoracic perimeter (TP), abdominal perimeter (AP), foreshank length (FSL), fore-shank width (FSW), fore-shank perimeter (FSP), tail width (TW), tail perimeter (TPe), hip width (HW), loin width (LWi), shoulder width (SW), forelimb length (FL), and body length (BL), were collected from 100 Corriedale ewes between 1.5 and 2 years old from the Illpa Experimental Centre of the National University of Altiplano in Peru. The machine learning algorithms used to estimate body weight were Support Vector Machines for Regression (SVMR), Classification and Regression Trees (CART), Random Forest (RF), Model Average Neural Networks (MANN), Multivariate Adaptive Regression Splines (MARS) and eXtreme Gradient Boosting (XGBoost). The performance of the models was evaluated by the coefficient of determination (R2), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). Highly correlated predictors (r >= 075) were removed from the dataset. The remaining predictors were then subjected to variable selection procedures using the Boruta algorithm. Boruta results confirmed the importance of TP, LWi, BL, FSL, SW and HW as predictors of ewe weight. The ML models were then trained on those selected predictors. RF had the highest R2 values and lowest values of MAE, RMSE, and MAPE. In conclusion, the RF algorithm can be recommended for accurately estimating BW from body measurements of Corriedale sheep.