Assessment of relationship between body weight and morphological traits of South African non-descript indigenous goats using different data mining algorithm

Abstract

Modern analytical techniques such as data mining algorithms are used to create a model that accurately estimates continuous dependent variable from independent variables of a given set of data. The present study used different data mining algorithms to assess the association between body weight (BW) and morphological characteristics such as body length (BL), heart girth (HG), withers height (WH), rump height (RH), and rump length (RL) of South African non-descript indigenous goats. The research was carried out in the Lepelle-Nkumbi Local Municipality, Capricorn District in the Limpopo province of South Africa. The study used 700 non-descript indigenous goats which include 283 bucks and 417 does with age ranged from one to five years old. The morphological characteristics were taken with a tailor measuring tape and a wood ruler calibrated in centimetres (cm), while the BW was taken with a balanced animal scale calibrated in kilograms (kg). Before the goats were allowed to go for grazing, the following body measurements (BW, BL, HG, WH, RH and RL) was taken once in the morning. Data was analyzed using descriptive statistics, Pearson correlation, various data mining algorithms (Chi-square automatic interaction detector, Classification, and regression tree), analysis of variance and goodness of fit equations (Coefficient of determination (R2), adjusted coefficient of determination (Ajd.R2), root mean square error (RMSE), relative approximate error (RAE), standard deviation ratio (SD. ratio) and coefficient of variance (CV)). The result showed that, BW and HG had higher mean values in does than bucks, BL and WH had higher mean values in bucks than does, and RH and RL had equal mean values in bucks and does, according to descriptive statistics. Furthermore, our findings showed that the BW of does had positive significant correlation (P < 0.01) with BL (r = 0.65), and positive significant correlation (P < 0.05) with HG (r = 0.28), but non-significant correlation (P > 0.05) with WH (r = 0.21), RH (r = 0.23) and RL (r = 0.23). However, the result for bucks indicated that BW had positive significant correlation (P < 0.01) with BL (r = 0.65) but non-significant correlation with HG (r = 0.22), WH (r = 0.07), RH (r = 0.14) and RL (r = 0.12). The chi-square automatic interaction detector and classification and regression tree results indicated that BL in bucks and does had statistical significance (P < 0.01) on BW followed by age, HG, and villages where the animals were raised. Goodness of fit results indicated there was high R2 = 0.58, Adj. R2 = 0.58, and low SD. Ratio = 0.65, RAE = 0.02, RMSE = 5.53) and CV = 14.49 in

CHAID model and low R2 = 0.51, Adj. R2 = 0.46 and high SD. Ratio = 0.70, RAE = 0.20, RMSE = 5.95 and CV = 15.49 in CART model. Analysis of variance results indicated that age had significant difference (P < 0.01) on BW and some morphological traits including BL, HG, WH and RH. Sex only revealed significant difference (P < 0.01) in RL. It was concluded that BL alone in both sexes can be used as a selection criterion when determining body weight of goats. Both CHAID and CART suggest that BL alone can be used as a predictor of body weight in goats. Goodness of fit calculations suggest that CHAID is the best model due to its high R2, Adj. R2 and low RAE and RMSE. Findings suggest that age can be used as deciding factor for the measured traits including BW, BL, HG, WH and RH in both does and bucks. Findings suggest that sex can only be used as a deciding for RL only in the current study.