An innovative mathematical approach for a highly informative treatment of automatic milking system datasets: development and testing of enhanced clustering models

New devices and studies have exponentially increased the amount of data available on modern farms. Farmers need to convert this data into useful, immediate and practical information for farm management and animal welfare. Therefore, in recent years many mathematical models already used in other fields, such as neural networks, image processing, regressions and clustering, have been applied to precision livestock farming. The goal of this study was to develop a clustering method for herd characterisation on dairy farms. Specifically, a cluster-graph approach was applied to a dataset collected through the automatic milking system (AMS) of an Italian dairy farm andcontaining real-time information for each cow: daily milking frequency, activity, parity, weight, milking frequency, and days of lactation. The clusters were updated every month within the study time span to reflect changes in animal conditions. The results represent a scientific method of transforming the amount of data available on the farm into user-friendly and highly informative data representations, such as graphs and plots, which can provide farmers with valid herd management support, particularly in terms of time and cost optimisation. By comparing subgroups for every single month, clustering can help farmers to take the most appropriate action promptly in order to increase animal welfare and productivity. Integration of the dataset into other farms through new devices and insights is planned in order to further refine the model.