Live Weight Prediction in Norduz Sheep Using Machine Learning Algorithms

The objective of this study was to compare predictive performances of four machine learning (ML) models: Support Vector Machines with Radial Basis Function Kernel (SVMR), Classification and Regression Trees (CART), Random Forest (RF) and Model Average Neural Networks (MANN) to predict live weight from morphological measurements of Norduz sheep (n=93). Seven morphological measurements; chest girth (CG), chest width (CW), chest depth (CD), height at withers (HW), body length (BL), heigth at rump (HR) and rump width (RW) were used to predict live weigth (LW) of Norduz sheep. All morphological measurements were positively correlated to LW. Live weight had the highest correlation with CG and the lowest correlation with HR. Initially, highly correlated predictors were removed from the data set. The remaining predictors were then subjected to variable selection procedures using the Boruta algorithm. The results of Boruta confirmed the importance of the four predictors HW, BL, CW, and CD. However, HR confirmed to be unimportant was excluded from the dataset. The ML models were trained on selected predictors. The results showed that the prediction performance validated using the test dataset indicated that RF had the lowest values of Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percent Error (MAPE). The permutation-based variable importance scores indicate that CW and CD were the most important variables in predicting LW. The actual LW had the highest significant positive correlations with the values predicted by SVMR and RF, and followed by ANN and CART models respectively. There were no differences between the means of actual and predicted LWs by machine learning models. The fact that the models generalized well on the testing data sets indicates that machine learning algorithms have valid predictive patterns and are effective methods in LW weight of Norduz sheep. Considering runtime of the models, although the CART model had the lowest computational cost, it had the worst performance. The MANN algorithm, on the other hand, required a longer runtime to process the same dataset.