Practical information

December 16, 2025, at 1:30 p.m. in amphitheater A-1 of building A on the AgroParisTech campus, 22 place de l’Agronomie in Palaiseau. 

Supervision

Thesis prepared under the supervision of Masoomeh TAGHIPOOR, Research Engineer (ADR), co-supervised by Christine DUVAUX-PONTER, Professor, and co-supervised by Nicolas C. FRIGGENS, Director of Research, and Joon KWON, Research Fellow.

Abstract

In the context of the agroecological transition of livestock systems, there is increasing societal and scientific demand for promoting livestock farming systems that are more respectful of animal welfare, while also supporting the selection of robust animals facing climatic and other environmental challenges. Precision Livestock Farming (PLF) technologies produce a wide range of time-series data, providing the continuous and individual monitoring of animals as affected by environment and husbandry practices. While PLF has advanced significantly in the detection of health and performance, its potential for dynamic behaviour monitoring at the individual level remains underexplored. This work aimed to address this gap by: (1) developing an automated and standardised tool to classify key behaviours from raw accelerometer data, designed to be applicable across experimental contexts and animal species, (2) using this tool to analyse individual behavioural activity profiles and to characterise behavioural deviations in response to an induced nutritional perturbation, and (3) improving the detection of short and/or infrequent behaviours, which are still difficult to detect using standard classification approaches but are essential for fine-scale welfare assessment. A pipeline (ACT4Behav) was developed, including flexible pre-processing and classification steps that enable the generation of continuous behavioural data. This pipeline was successfully applied in different contexts, demonstrating robust performance across individuals and sensor placements.

The use of resampling techniques improved the performance scores of the models for the detection of headbutt, a short and infrequent behaviour. Moreover, the pipeline was used to produce individual behavioural data and assess their sensitivity to a nutritional challenge. The results showed marked inter-individual variability in behavioural flexibility. Whilst some goats exhibited clear changes in their activity patterns in response to the perturbation, others maintained relatively stable profiles. This high variability among goat responses suggests the existence of distinct coping strategies among individuals. Quantifying such variability is essential not only for improving welfare monitoring but also for guiding management and breeding decisions in the context of changing environments caused by climate change. This PhD project contributed to the development of tools for individualised behaviour monitoring and welfare assessment. It highlights the importance of accounting for individual variability in behavioural responses and paves the way for integrating welfare indicators into precision livestock farming strategies.