Artificial Neural network and regression models to predict BW of Holstein and Brown Swiss cattle using real-time body measurements

In this study, it was aimed to evaluate the possible use of some morphometric body measurements to predict body weight (BW) of Holstein (H) and Brown Swiss (BS) cattle grown in a 12-month Intensive beef production system. Associations between bodyweight (BW) and some body measurements such as heart girth (HG), wither height (WH), body length (BL), body depth (BD), hip width (HW), hip height (HH) and pin bone length (PL) were examined for prediction ability, using the data with 1068 observations for each trait. The linear, quadratic and cubic regression models (REGR) were performed to predict BW for both breeds and since there were no statistically significant differences in body measurements between breeds. The data of these breeds were combined and found that BL and HG would be the most advisable traits in estimating BW (R2=95.8% and 95.7% respectively) among the other measurements. Regression analysis showed that addition of quadratic and cubic terms provided little contribution in BW prediction. Therefore, linear terms of all variables were considered for analysis and they were significant and R2 values for other body measurements WH, HW, BD, HH and PL were approximately 93.2, 79.9, 87.1, 90.9 and 79.2% respectively. The best predictors were the same to estimate BW by Artificial Neural Networks (ANN). The best predictive variables were HG and BL by ANN with R2= 98.2 and 98.1% respectively. In conclusion, BW can be estimated precisely by both REGR and ANN using BL and HG alone or both HG and even WH and different models may be needed to predict BW in different feeding and environmental conditions and breeds. However, ANN produced higher accuracy in prediction of live weights than REGR.