Efficacy of Using the Elastic net Regularized Regression Method to Predict Body Weight of Beef Heifers Using Body Measurements and Calculated Frame Score

Abstract Several methods to estimate the body weight (BW) of cattle have been proposed. Body measurements have proved to be useful in estimating the BW of cattle. However, there are multicollinearities between body measurements which could affect the accuracy of prediction models. Least absolute shrinkage and selection operator (LASSO regression), ridge regression, and more recently elastic net (EN) regularization has been suggested to handle multicollinearity better than simple regression. Elastic net is a regularization technique that combines LASSO and ridge regression. In this study, we modelled BW against some body measurements [calculated: volume (VL) 218 to 503 liters; frame score (FS) 2.1 to 7.5; surface area (SA) 4.02 to 7.1 m2], and direct body measurements [body length (BL) 89 to 136 cm; hip height (HH) 96 to 136 cm; hip width (HW) 33 to 52 cm; heart girth (HG) 149 to 191 cm; mid girth (MG) 177 to 234 cm; flank girth (FG) 156 to 239 cm] to examine which variable(s) predicts BW more accurately. Data from 300 heifers (9 breeds) collected over 4 years were used with year as a random effect to account for differences between years. We ran the EN model using the glmnet function of the R caret package. Ten-fold cross validation was used with 5 repeats. The EN model was trained using 206 observations and validated with 94 observations. The best model was selected based on the minimum root mean square error and median R-square (R2) values. Including FS in the prediction model resulted in the highest R2 (0.5754) and was a better predictor for BW compared with other models: VL (0.4271), SA (0.4130) and direct body measurements (0.4861). This suggests that calculated FS could be a better estimate for BW than direct body measurements and indicates that FS still accurately predicts size of heifers.