Use of Automatic Linear Modeling and Decision Trees in Body Weight Prediction in Normal Feathered Chickens

Abstract

The aim of the study was to predict body weight (BWT) from linear body measurements of 200 mature normal feathered chickens, comprising 150 males and 50 females randomly selected in Amansea, Awka North LGA, Anambra State, Nigeria. Automatic Linear Modeling (ALM) and decision tree algorithms involving Chi-square Automatic Interaction Detector (CHAID), Exhaustive CHAID, and Classification and Regression Trees (CART) were used in the prediction. Adjusted coefficient of determination (R²adj.) and Alkaike’s Information Criterion corrected (AICc) were used to evaluate the predictive ability of ALM in males, females and pooled sexes. Males were significantly (p<0.05) superior to females in BWT (1.26±0.02 kg vs 1.0.5±0.03 kg) and many linear parameters. The ALM indicated that body length (BL), shank length (SL) and breast width (BW) had the highest significant (p<0.001) fractional importance in BWT prediction in males (0.793), females (0.721) and pooled sexes (0.501), respectively. Highest predictive ability of ALM was achieved in pooled sexes with the smallest AICc (-0.572.92), and   predicted BWT value of 1.49 kg. CHAID and Exhaustive CHAID revealed that highest BWT of 1.220 kg could be predicted with BW>9.000 cm. With CART, no variable was identified as important in BWT prediction. The study revealed ALM as the best model for predicting BWT using BW of pooled sexes in normal feathered local chicken.