Final weight prediction from body measurements in Kıvırcık lambs using data mining algorithms

This study was carried out to determine the final weight estimation of Kıvırcık lambs using body measurements via Chi-square automatic interaction detection (CHAID), exhaustive CHAID, classification and regression tree (CART), random forest (RF), multivariate adaptive regression spline (MARS), and bootstrap-aggregating multivariate adaptive regression spline (Bagging MARS) algorithms. For this purpose, height at withers (HW), back height (BH), croup height (CH), chest depth (CD), body length (BL), chest width (CW), and chest circumference (CC) were measured in the lambs. The statistical performances of these algorithms (CHAID, exhaustive CHAID, CART, RF, MARS, and Bagging MARS) were tested by using several goodness-of-fit criteria, namely the coefficient of determination ( $R^2=0.699$ , 0.699, 0.722, 0.662, 0.792, and 0.624), adjusted coefficient of determination (Adj. $R^2=0.633$ , 0.633, 0.721, 0.637, 0.768, and 0.609), coefficient of variation (CV % $=$ 6.35 and 5.14, $P<0.01$ ), mean square error (MSE $=$ 3.296, 3.296, 2.904, 4.461, 2.277, and 4.121), root mean square error (RMSE $=$ 1.815, 1.815, 1.704, 2.112, 1.509, and 2.030), mean absolute error (MAE $=$ 1.409, 1.409, 1.279, 1.702, 1.193, and 1.628), and mean absolute percentage error (MAPE $=$ 3.925, 3.925, 3.578, 4.002, 3.335, and 3.967), between actual and predicted values of live body weight. With this, the best-fitted MARS model was chosen using cross-validation and user-defined parameter optimization. As a result, it has been shown that it is possible to make a successful estimation of the live weights of lambs by using some of the morphological features of the lambs.