Comparison of data mining algorithms and regression method for prediction of live body weight using biometric traits of non-descript indigenous chickens

Abstract The study aimed to investigate the relationship between live body weight and biometric traits, and to estimate live body weight from biometric traits of non-descript indigenous chickens. A total of 400 non-descript indigenous chickens (113 males and 287 females) were randomly selected from a multi-stage purposive sampling procedure following a cross-sectional study design. Live body weight (LBW), body length (BL), beak length (BKL), shank length (SL), shank circumference (SC), and comb height (CH) were measured. Data was analysed using Pearson’s correlation, CHAID, CART, and stepwise linear regression. Correlation results revealed that LBW of males had a positive and high correlation ( P < 0.01) with BL ( r = 0.63), and LBW of females had a significant positive correlation ( P < 0.01) with SC ( r = 0.45). Regression results showed that for males, model combining BL, BKL, CH and SC (R 2 = 0.62, RMSE = 15.26, AIC = 2183.27, BIC = 2199.23) was best in estimation of LBW, while for females, model combining SC, BL, BKL and CH (R 2 = 0.36, RMSE = 12.31, AIC = 2011.20, BIC = 2027.17) was best in estimation of LBW. CHAID results revealed that SC was the best variable to estimate LBW, while CART model revealed that sex was the best variable for estimation of LBW. CHAID model ( r = 0.79, R 2 = 0.63, RMSE = 11.73, AIC = 1969.45, BIC = 1973.44) outperformed stepwise regression and CART models. The study concludes that there was no model that had a good performance to predict live body weight of the chickens.