Feature selection and intelligent livestock management

Computational animal breeding relies on genetic-statistical models that are aimed at estimating breeding values, which in turn are used to rank animals based on their genetic potential with respect to certain traits of interest (e.g. size, milk yield). However, modern livestock production systems collect large amounts of data throughout the life of an animal that are not directly suited for those statistical models, such as periodic phenotype values (physical traits such as weight measurements) and environmental observations. In this thesis, we explore the potential of using that additional data to improve future phenotype prediction in livestock using machine learning methods.