Use of MARS Data Mining Algorithm Based on Training and Test Sets in Determining Carcass Weight of Cattle in Different Breeds
No Thumbnail Available
Date
2022
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Ankara Üniversitesi
Abstract
This research was carried out with the purpose of estimating hot carcass weight by using parameters such as race, carcass weight and age with Multivariate Adaptive Regression Spline (MARS) algorithm. To achieve this goal, 700 cattle data belonging to the years 2017-2018, which were taken in equal numbers from 7 different breeds, were used. A total of 700 data were used, taking equal numbers of data from each breed. In order to test the accuracy of the model created in the research, the data set was divided into two data subsets as training and test subsets. In order to test the compatibility of these separated subsets with the MARS model, a new package program named “ehaGoF” which estimates 15 goodness of fit criteria was used. According to the analysis results, the MARS model with the smallest SDRATIO (0.157, 0.130) and the highest determination coefficient (R2) (0.975, 0.983) of the training and test sets, respectively, was determined. Looking at the other fit values, it is seen that the training and test set are quite compatible. In terms of hot carcass weight among the breeds, it was determined that the Limousine race performed higher than the other breeds. As a result, the implementation of the MARS algorithm can allow livestock breeders to obtain effective clues by using independent variables such as breed, age, and body weight in estimating hot carcass weight.
Description
Keywords
K-fold Cross Validation, Carcass weight, ehaGoF