Heterogenous, multisource and multidimensional data mining to develop machine learning algorithms for early prediction of lameness in cattle

This exciting opportunity is based within Ruminant Population Health group at the School of Veterinary Medicine and Science which conducts cutting-edge research into the health and welfare of UK cattle and sheep, and is founder members of the national ‘Centre for Innovation Excellence in Livestock’ (CIEL) with recently opened ‘Centre for Dairy Science Innovation’. Industry funder and partner is CRV (https://www.crv4all-international.com/) an innovative herd improvement cooperative that offers dairy and beef farmers worldwide a wide-ranging portfolio of genetics, herd management products and services. Over the last 150 years CRV has collected data on millions of cows and thousands of farms and our data platform is rapidly expanding.

Lameness is one of the most important endemic diseases present in cattle around the world in terms of both animal welfare and economic loss. Currently, tools to identify lameness rely on visual assessment of the cow using a subjective scoring scale and there are no predictive algorithms that can identify lameness early. Precision livestock tools have enabled us to capture, store and process large amounts of data on farm and this PhD is unique with data availability from thousands of dairy farms for many years. Temporal data that is available has been collected on farms via sensors and alternative methods that will provide novel information about individual cow behaviour, genetics, claw health, milk recording, fertility etc. These data will be used to develop algorithms for lameness. While recent advances in artificial intelligence and machine learning techniques have boosted the potential for analysing such ‘big data’ to develop predictive algorithms, there are a few key challenges: data heterogeneity in terms of type and frequency of data and feature selection and accuracy of algorithms.

This industry linked interdisciplinary PhD project thus aims to use a range of methodologies from across disciplines of veterinary science (lameness epidemiology, precision livestock, animal behaviour), statistics (interpolation methods such as spine interpolation and gaussian process regression, shrinkage) and computer science (machine learning especially deep learning) to overcome above challenges and create new knowledge and tools to predict lameness in dairy cows.

The aim of this interdisciplinary PhD project is to utilise large amounts of heterogenous data collected on farms by CRV (thousands of farms for 5 years) to:
• Develop and compare algorithms using supervised, unsupervised and semi-supervised machine learning methods that can predict claw health problems in cattle
• Validate the developed algorithms in the field by collecting new data

The project will be based at the School of Veterinary Science with some time at the School of Mathematical Science where the student will benefit from interaction with a thriving community of postgraduate students and postdocs within two schools.