Machine Learning Model for Maternal Quality in Sheep

Abstract

This paper aims to identify determinant traits of ewes by measuring their impact on lamb survival. For that, we devised a machine learning model that correlates ewe traits to lamb survival, and figured out as to which ewe traits explain the correlation and hence help us to identify the better mother. In this study, we kept pregnant ewes under 24 h observation by two researchers starting approximately three days before expected parturition dates. We conducted the study using native and crossbreed lambs produced in high altitude and cold climate region. It is critical to note that parturation took place with minimum interruption unless there is a birth difficulty. Independent variables used in the machine learning model pertain to mother's behaviours during parturation, however, we also took into consideration factors like dam breed, dam body weight at lambing, age of dam, litter size at birth, lamb breed and sex. Lamb survival is a nominal output variable, hence we tried out several classification algorithms like Bayesian Methods, Artificial Neural Networks, Support Vector Machine and Tree Based Algorithms. Classification algorithms applied for lamb survival were Bayesian Methods, Artificial Neural Networks, Support Vector Machine and Trees. RandomForest algorithm was found best performer among tree algorithms. We were able to present tree visualisation for mothering ability with 80% accuracy rate and 0.43 Kappa Statistics. The result of the study shows that grooming behaviour is the first determinant mothering ability. If the grooming duration is longer than 15 minutes, then it is a good mother.