ABSTRACT
This paper aims to predict male and female camels' mature weight (MW) through various morphological traits using hybrid machine learning (ML) algorithms. For this aim, biometrical measurements such as birth weight (BW), length of face (FL), length of the neck (NL), a girth of the heart (HG), body length (BL), withers height (WH), and hind leg length (HLL) were used to estimate the mature weight for eight camel breeds of Pakistan. In this study, multivariate adaptive regression splines (MARS), random forest (RF), and support vector machine (SVM) were applied to develop prediction models. Furthermore, the artificial bee colony (ABC) algorithm is employed to optimize ML models' internal parameters and improve prediction accuracy. The predictive performance of ML and hybrid models was evaluated on a testing dataset using goodness-of-fit measures such as mean absolute deviation (MAD), mean absolute percentage error (MAPE), coefficient of determination (R2), and root mean square error (RMSE). The results of the study revealed the ABC-SVM model was the best predictive model. The experimental results of this study showed that the proposed ABC-SVM method could effectively improve the accuracy for MW prediction of camels, thus having a research and practical value.
Subject(s)
Algorithms , Camelus , Male , Female , Animals , Machine Learning , Biometry , Random ForestABSTRACT
Mature weight is a significant trait that can be influenced by age, sex, breed, production system, and climate conditions in camels. In camel breeding, it is essential to describe breed standards of the studied camel breeds as part of morphological characterization and to determine morphological traits positively influencing mature weight within the scope of indirect selection criteria. This study was to find the best one among candidate models in prediction of mature weight from several morphological traits measured for eight camel breeds (Bravhi, Kachi, Kharani, Kohi, Lassi, Makrani, Pishin, and Rodbari) raised under Pakistan conditions. The morphological measurements taken from the camels in the study were birth weight (BW), weaning weight (WW), mature weight (MW), age of ridding (ARD), face length (FL), face width (FW), head length (HL), head width (HW), ear length (EL), ear width (EW), neck length (NL), neck width (NW), hump length (HL), hump width (HuW), heart girth (HG), withers height (WH), body length (BL), fore leg length (FLL), and hind leg length (HLL), respectively. In the prediction of mature body weight as a response variable, the optimal MARS predictive model with 15 terms selected by train function of the caret package produced very high predictive performance without encountering overfitting problem. Goodness of fit criteria were estimated to measure predictive quality of the MARS model using ehaGoF package available in R environment. Morphological characterization of the camel breeds was performed with hierarchical cluster analysis (HCA) on the basis of Euclidean distance-Single linkage. At the first step of hierarchical cluster analysis, the similarity level of Bravhi and Kachi camel breeds was the highest with 85.3569 (%). At the second step, Makrani joined to new cluster of Bravhi and Kachi camels found at the first step, and the similarity level of the new cluster comprising Bravhi, Kachi, and Makrani breeds was found as 84.5562 (%). MW was significantly correlated with BW (0.677), WW (0.536), HL (0.524), HuW (0.529), and ARD (0.375) at P < 0.01, and there was the highest correlation of 0.994 between HHL and FLL (P < 0.01). As a result, it could be suggested that results of MARS modeling may help camel breeders to reproduce the elite camel populations and to describe characteristics associated positively with MW within the scope of indirect selection criteria.
