Application of Multivariate Adaptive Regression Splines (MARS) for Modeling the Lactation Curves


ORHAN H. , Teke E. C. , Karci Z.

KSU TARIM VE DOGA DERGISI-KSU JOURNAL OF AGRICULTURE AND NATURE, vol.21, no.3, pp.363-373, 2018 (Journal Indexed in ESCI) identifier

  • Publication Type: Article / Article
  • Volume: 21 Issue: 3
  • Publication Date: 2018
  • Doi Number: 10.18016/ksudobil.334237
  • Title of Journal : KSU TARIM VE DOGA DERGISI-KSU JOURNAL OF AGRICULTURE AND NATURE
  • Page Numbers: pp.363-373

Abstract

The aim of this study is to model milk yield using the MARS method using independent variables such as Holstein cows control day, milking time, conductivity and mobility. MARS is a non parametric method for predicting linear sub models to determine appropriate knot points of non linear models. This study included daily lactation records for 80 Holstein cows between 2006 and 2011. For each lactation, the most suitable model was determined by testing different maximum interaction models. The model suitability is generally assessed by the criteria that generalized cross validation criterion (GCV) minimum and R-2 maximum values. When these criteria are taken into consideration, the non interactive model for the first, four lactations and the 3 interacting model for the fifth lactation are determined as the best models. The determination coefficients (R-2) of the MARS models according to the lactation order are found to be 0.983, 0.991, 0.991, 0.975 and 0.950, respectively. All the independent variable coefficients in models were found to be important at 99% level. In all models, MARS has been identified as the most meaningful variable of control day. According to these results, we can say that the estimation of milk yield of models produced by MARS is successful and safe.