Machine learning algorithms for lamb survival
Lamb survival is influenced by the culmination of a sequence of often interrelated events including genetics, physiology, behaviour and nutrition, with the environment providing an overarching complication. Machine learning algorithms offer great flexibility with regard to problems of complex interactions among variables. The objective of this study was to use machine learning algorithms to identify factors affecting the lamb survival in high altitudes and cold climates. Lambing records were obtained from three native breed of sheep (Awassi = 50, Morkaraman = 50, Tuj = 50) managed in semi intensive systems. The data set included 193 spring born lambs out of which 106 lambs were sired by indigenous rams (n = 10), and 87 lambs were sired by Romanov Rams (n = 10). Factors included were dam body weight at lambing, age of dam, litter size at birth, maternal and lamb be-haviors, and lamb sex. Individual and cohort data were combined into an original dataset containing 1351 event records from 193 individual lambs and 750 event records from 150 individual ewes. Classification algorithms applied for lamb survival were Bayesian Methods, Artificial Neural Networks, Support Vector Machine and Decision Trees. Variables were categorized for lamb survival, lamb behavior, and mothering ability. Random-Forest performed very well in their classification of the mothering ability while SMO was found best in predicting lamb behavior. REPtree tree visualization showed that grooming behavior is the first determinant for mothering ability. Classification Trees performed best in lamb survival. Our results showed that Classification Trees clearly outperform others in all traits included in this study.