Investigation into the use of sensor and production data to predict lameness in dairy cows

Hemingway-Arnold, Heather (2023) Investigation into the use of sensor and production data to predict lameness in dairy cows. MRes thesis, University of Nottingham.

[img] PDF (Thesis - as examined) - Repository staff only - Requires a PDF viewer such as GSview, Xpdf or Adobe Acrobat Reader
Available under Licence Creative Commons Attribution.
Download (2MB)


The object of this research was to investigate the effect of lameness on individual sensor and production parameters and combine this data to predict lameness in dairy cows. Lameness is an extremely important welfare issue currently facing the worldwide dairy industry. Traditional lameness detection by mobility scoring is subjective, variable and has low sensitivity: rendering it an insufficient tool in combating the high UK dairy lameness prevalence. Automatic lameness detection is undergoing constant investigation to support early lameness detection, but currently there is no system that offers an accurate and practical solution for implementation on farm. This study used a neck-mounted accelerometer, leg-mounted accelerometer and an automatic milking system to collect a plethora of data of 110 cows over a 3-month period. Mobility scoring occurred twice-weekly using an adapted version of the AHDB mobility score (0 = Perfect mobility, 1 = Imperfect mobility, 2A = Mildly Lame, 2B = Moderately Lame, 3A= Severe Lame, 3B = Non-Weight Bearing). There were considerable differences observed between lying time, activity, average weight and parity. Four random forest models were constructed using 1. ICEQube data (lying time and activity), 2. Lely Qwes-H data (activity and rumination), 3. Production data and 4. All data combined. The combined model achieved the best sensitivity, and specificity, at 0.74 and 0.75, and AUC at 0.82. This demonstrates that combining data from multiple sources improves predictive accuracy, although care must be taken when including confounding variables such as weight and parity. Model performance was improved when detecting severely lame cows over mildly lame cows. Combining multiple sensor technologies shows promise in improving detection of lameness in dairy cows, although challenges such as generalisability and variability must be overcome to improve sensor performance.

Item Type: Thesis (University of Nottingham only) (MRes)
Supervisors: Green, Martin
O'Grady, Luke
Clifton, Rachel
Keywords: lameness, dairy cows, sensor technologies
Subjects: S Agriculture > SF Animal culture
Faculties/Schools: UK Campuses > Faculty of Medicine and Health Sciences > School of Veterinary Medicine and Science
Item ID: 72070
Depositing User: Hemingway-Arnold, Heather
Date Deposited: 31 Jul 2023 04:40
Last Modified: 31 Jul 2023 04:40

Actions (Archive Staff Only)

Edit View Edit View