Comparison of selected decision tree algorithms in the prediction of body weight in awassi lambs

Abstract

The present study was conducted to evaluate the comparative predictive performances of Classification and Regression trees (CART), Chi Squared Automatic Interaction Detector (CHAID) and Exhausted CHAID algorithms used to predict body weights of Awassi lambs at 60-d (W60) and 90-d (W90) of age. For this purpose, 730 Awassi lamb records were collected from 3 base flocks in Turkey in 2014-2016. The potential predictors included in this study were dam age, sex, birth type, flock, lambing season and birth year. In order to determine the best one among these decision tree algorithms, model evaluation criteria i.e RMSE, MAPE, RAE, SDratio, MAD, Pearson coefficient, Coefficient of determination (R2) and adjusted coefficient of variation (R2Adj) values were calculated. For the prediction of W60 and W90, the best decision tree algorithm was found to be the CART algorithm. R2 for W60 and W90 were 0.614 and 0.978 and RMSE estimates for W60 and W90 were 0.94 and 0.321, respectively. The influential predictors affecting W90 were flock and W60. However, flock, birth weight (BW), birth type and birth year were found as significant factors for W60. In conclusion, CART algorithm may be a useful tool in describing breed standards of the Awassi for selection purposes in animal breeding. Also, it outperformed Exhausted CHAID and CHAID decision tree algorithms in predictive performance to predict W60 and W90 of Awassi lambs.