Live Weight Prediction in Norduz Sheep Using Machine Learning Algorithms

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