[PhD student] Adrien Cotil: Development of a mathematical method based on proximity data sensors for the early detection of pathologies in farm animals

Development of a mathematical method based on proximity data sensors for the early detection of pathologies in farm animals

My name is Adrien Cotil and I'm currently doing an internship at Inrae in Montpellier, as part of my Master 2 in "Probability and Random Models" at Jussieu University (Paris 6). After a preparatory class in Biology (BCPST), I wanted to move into research in mathematics applied to biology. The idea of this thesis is to establish an algorithm capable of detecting pathologies in farm animals (such as mastitis or lameness) from high-frequency data on animal movements, based on two hypotheses: that the social network of inter-individual spatial proximities possesses a certain temporal stability, linked to individual characteristics, as a result of the different events induced by farming practices, and that health disorders or damage to the well-being of individuals can modify their position in this network in a way that is early and detectable by adapted algorithms.

The thesis will be in two parts: the first will present a study of the different ways of modeling the movement of farm animals in the literature and will propose a variant of these models adapted to our context. The second part will use the previously established model(s) to design break detection algorithms enabling the automatic detection of disturbances in the social organization of the herd (in order to identify the individuals involved) and the identification of disturbances in the individual behavior of each animal (in order to prevent incipient pathologies).

I chose this subject because I'm particularly sensitive to animal welfare and, more generally, to issues arising from agronomy and ecology. I think this project will be useful to both animal behavior researchers and veterinarians when they work on farms.

• Starting date : 1st November 2021
• University : Université de Montpellier
• PhD school : I2S – Information Structure Système
• Scientific field : Mathematics, IT
• Thesis management : Bertrand Cloez, UR Mistea, Inrae
• Thesis supervisors: Jean-Baptiste Menassol, UMR Selmet, Institut Agro
• Funding: #DigitAg – Inrae
• #DigitAg : Cofunded thesis – Axe 6 : Modélisation et simulation (systèmes de production agricole), Axe 2 : Innovations en agriculture numérique, Axe 3 : Capteurs, acquisition et gestion de données, Challenge 4 : Des productions animales durables

Keywords: Process statistics, Clustering - break detection, Stochastic modeling, Individual-centered model, Zootechnics, Animal behavior

Abstract: Animal production is one of the agricultural sectors with a heterogeneous operational deployment of ICTs. Ruminant breeding is a good example of this contrast with sectors such as dairy cattle, which are well supplied, and sectors with a more modest level of commercial digital equipment, such as suckling sheep. While there is no lack of sensors and associated tools, it is more a question of the relevance and reliability of the information that is derived from them, thanks to sufficiently precise and predictive algorithms that still need to be developed. The use of these new sensors creates the need for new methodologies, such as the modeling of social interactions from spatio-temporal data including advanced statistical inferences, to produce more accurate predictive information in real time. We propose to work on an interdisciplinary approach, using mathematical modeling, inferential statistics and the study of social behaviors. The aim of the thesis is to adapt these approaches to ruminant (cattle and sheep) movement data, in order to identify breaks in the social structure of a group of farm animals that could be early indicators of individual pathologies. For this purpose, the PhD student will set up new non-parametric statistical estimators for a new model of diffusion process interactions (based on existing models) that will allow the production of specific algorithms for clustering and/or detection of disruptions.