Abstract: This study utilized a model based on the least square support vector machine (LSSVM) for the prediction of body weight (BW) of sheep. Two heuristic algorithms, namely, coupled simulated annealing (CSA) and simplex method (S) were applied to tune the hyperparameters of the LSSVM model. The hybrid CSA-S-LSSVM method is then applied to the male, female and total data of the Harnai sheep breed of Pakistan. Various biometric traits such as withers height, body length, chest girth, face length, paunch circumference, ear length, the length between ears, fat tail width and fat tail length were used as predictors. Goodness of fit measures such as mean absolute error (MAE), root mean square error (RMSE), adjusted coefficient of determination (Adj.R$^{2}$), normalized mean square error (NMSE) and mean absolute percentage error (MAPE) were used for evaluation. Comparison of the predictive performance of the proposed model on 10-fold cross-validation against both conventional (ridge regression) and state of the art machine learning (artificial neural networks) methods showed that the CSA-S-LSSVM outperformed both competing models by achieving the least values for MAE, RMSE, NMSE and MAPE and the highest value for Adj.R$^{2}$ in both training and testing data sets. The results were promising, accurate and viable for the prediction of BW of sheep.

Keywords: Body weight, LSSVM, artificial neural networks, ridge regression, body measurements, deep learning

Full Text: PDF