Allyndrée Joseph

After a double Master of Science in data science (AgroParisTech - IODAA) and bioinformatics (Université Paris-Saclay - AMI2B), Joseph completed an internship in Dublin to extract calf behaviors from accelerometer and video data (the exact title was: "Classification of calf behaviors from accelerometer data using Machine Learning methods"). As the internship was very interesting and the subject was more complex than it first appeared, he decided to continue with a similar study as part of his PhD candidature. This PhD, supervised by Aurélien Madouasse (INRAE, UMR Bioepar), Antoine Cornuéjols (INRAE, UMR MIA) and Christine Martin (INRAE, UMR MIA) is entitled: "Development of methods for characterizing the behavior and social interactions of cattle for health and welfare assessment in the context of agro-ecological transition". It will be divided into three distinct parts:

1. Data collection on dairy farms.

• Cattle fitted with collars containing sensors
• Recording of cattle behaviors: rising, lying down, other usual behaviors interactions; direct or video observation
• Labeling of lameness data using a standard scoring method.

2. Evaluation of machine learning methods for predicting observed behaviors from sensor data.

• Accelerometric signal: methods based on segmentation of time series into series into overlapping windows, followed by the application of adapted algorithms, or deep learning methods to identify specific patterns.
• Proximity detection: approach to be developed and methods to be identified.

3. Evaluation of associations between behaviors, social interactions and lameness.

Start of thesis: October 2023

Key words: Machine learning, time series classification, animal health, lameness.