Detecting dairy cows' lying behavior using noisy 3D ultra-wide band positioning data

In precision livestock farming, technology-based solutions are used to monitor and manage livestock and support decisions based on on-farm available data. In this study, we developed a methodology to monitor the lying behavior of dairy cows using noisy spatial positioning data, thereby combining time-series segmentation based on statistical changepoints and a machine-learning classification algorithm using bagged decision trees. Position data (x, y, z -coordinates) collected with an ultra-wide band positioning system from 30 dairy cows housed in a freestall barn were used. After the data preprocessing and selection, statistical changepoints were detected per cow-day (no. included = 331) in normalized 'distance from the center' and (z) time series. Accelerometer-based lying bout data were used as a practical ground truth. For the segmentation, changepoint detection was compared with getting-up or lying-down events as indicated by the accelerometers. For the classification of segments into lying or non-lying behavior, two data splitting techniques resulting in 2 different training and test sets were implemented to train and evaluate performance: one based on the data collection day and one based on cow identity. In 85.5% of the lying-down or getting-up events a changepoint was detected in a window of 5 minutes. Of the events where no detection had taken place, 86.2% could be associated with either missing data (large gaps) or a very short lying or non-lying bout. Overall classification and lying behavior prediction performance was above 91% in both independent test sets, with a very high consistency across cow-days. This resulted in sufficient accuracy for automated quantification of lying behavior in dairy cows, for example for health or welfare monitoring purposes