Abstract:The development of modern dairy farming is aimed at creating larger farms with more intensive production in accordance with the market needs. New larger livestock facilities need new approaches to management and logistics. The Internet of Things (IoT), machine learning (ML) and cloud computing are forming the basis of digital animal husbandry as they are increasingly being introduced into dairy farming. The collection of IoT sensor data and their storage in the information cloud allows the use of machine learning methods for predicting events in livestock farms. ML is characterized by its hunger for computing resources in all its phases, which can be resolved using cloud computing. Collected data by IoT devices require cleaning and scaling. Building a model requires training, testing, and validation. All these activities should be carried out in a timely sequence. The purpose of this article is to build models trained to predict the future amount of milk with the greatest accuracy for each individual animal. To achieve this goal, a pure Azure Cloud DataFlow (ADF) has been created, which monitors the processes from collecting and storing Io� data, to data processing, modelling, and model evaluation to visualization of results. Following this data flow, the experimental studies described in this article are performed. Three regression machine learning models were trained on the data collected from a Smart Livestock farm. Testing of the developed models has proven the applicability of the developed Cloud DataFlow, as the Boosted Decision Tree Regression Model shows the highest accuracy in predicting the amount of milk produced by each individual animal.

CLOUD DATAFLOW FOR MACHINE LEARNING MODELING ON IOT DATA IN SMART LIVESTOCK FARMING