"LiveEdge"

"This dataset contains tri-axial accelerometer recordings from wearable ear-tag sensors (MetaMotionC) attached to two pigs at the Virginia Tech Swine Facility. Data were collected at 50 Hz over 10 nonconsecutive days, yielding 2,276,450  raw samples. Sensor streams are segmented into 273,108 windows of 1.5 s duration with 50% overlap. Each window is annotated with one of six behavior labels: Lying (39.1%), Eating (32.0%), Standing (10.7%), Walking (10.3%), Drinking (2.4%), and Interacting (5.6%). Ground-truth labels were obtained via synchronized RGB video (30 fps) and manual annotation. The dataset is provided as NumPy arrays: windows.npy (N, 75, 6) containing 3-axis accelerometer and gyroscope signals, and labels.npy (N,) containing string behavior labels. Pre-subsampled versions at 25, 12.5, 10, and  6.25 Hz are also included for adaptive sampling rate research. This dataset supports research in energy-efficient livestock behavior recognition, adaptive sensing, and edge intelligence for IoT applications."