Simulation de comportement d’une vache à l’aide d’un processus markovien

Dairy cow activity data is collected hourly to detect behavioural anomalies that could be precursors to health problems. This data is supplemented by tags indicating the cow’s health status. These are collected daily by the farmer. Cows behave differently when they are sick. However, due to the possibility of labelling errors, it is difficult to assess the robustness of an anomaly detection algorithm. We propose to build a simulation model of dairy cow behaviour using Markov chains. The aim is to create a synthetic dataset close to reality on which it will be possible to test different anomaly detection algorithms. Fuzzy rules will also be introduced to represent a data set where the transition from one health state to another is gradual.