Research papers of the week – September 5, 2022

Jarosław Wełeszczuk, Barbara Kosińska-Selbi, Paulina Cholewińska

Livestock Science

Ministerial score = 140.0
Journal Impact Factor (2022) = 1.929 (Q2)

Models built using machine learning algorithms (MLA) have been used to handle numerous challenges in various farming systems. In this study we wanted to propose an alternative approach of predicting the economical index (EI) and calving interval (CI). The original dataset was generated by Center of Genetics and it was collected by using customised python script with implemented Beautifulsoup module (version 4.9.3) and selenium module (version 3.141.0). The dataset described 1000 best Polish Holstein heifers and 1000 best Polish Holstein cows. We focus on modelling the predictions of EI and CI based on regression methods, by implementing different white-box models algorithms: linear regression (LR), Decision tree (DT), K-nearest neighbour's (KNN) and black-box models algorithms: random forest (RF), gradient boosting (GB), extreme gradient boosting regressor (XGB), neural network (NN). The heifer/cow dataset had to be split into 5 parts, by using K-Fold-Cross-Validation (K-Fold CV) strategy. Based on Recursive Feature Elimination with Cross-Validation (REFCV) we selected the best subset of features for each of the estimators. The best representation of the dataset for the regardless MLA was chosen based on different feature selection and feature engineering techniques described in the manuscript. Results obtained of mean absolute error (MAE) and root mean square error (RMSE) showed that the best predictions of EI were obtained with a model built using NN MLA (MAE: 20.72; RMSE: 29.35). The best predictions of CI were obtained with a model built using GB MLA (MAE: 0.79; RMSE: 1.27). Proposed methods of prediction of EI and CI could be useful for dairy cattle breeders for forecasting economical information without the need of long-term observations of individuals.

DOI:10.1016/j.livsci.2022.105039

READ THE PAPER UPWr Base