Automated Classification of a Calf’s Feeding State Based on Data Collected by Active Sensors with 3D-Accelerometer

The paper deals with the problem of time series classification for the feeding state of calves by means of features evaluated for acceleration real-time data sets. The eartags equipped with an active sensor were developed for location and animal activity identification. Video records synchronized with a sensor data were collected from three calves. After the data preprocessing including the reconstruction of lost information, filtering and frequency stabilization, new time series were used to develop a machine-learning algorithm with equidistant and non-equidistant time series segmentation method based on a modified Kolmogorov-Smirnov statistic. The proposed classification method has achieved a good recognition quality for the feeding state with a best overall accuracy of approximately 94%. Thus this methodology is useful in identifying the feeding state and we may expect the possibility to generalize it to the multi-state case as well. The further improvement of the algorithm is a subject of our future research.