Cows' behavior classification using acceleration data: A new, effective, and simple approach

Abstract

Monitoring and classifying cow behaviors provides valuable support for livestock management. This can be done through sensors attached to the pet. Due to their small size, light weight, and high accuracy, accelerometers are well-suited for this purpose. However, the complexity of behaviors, which often involve similar movements, poses challenges in interpreting the sensor data. This paper presents a novel classifier design for cow behaviors based on acceleration data and a specific set of features. By analyzing cow acceleration data, we extracted features for classification with the help of machine learning algorithms. With five features—Mean, Standard Deviation, Root Mean Square, Median, and Range—and a 15-second data window (1 sample/second), the classifier achieved optimal performance when identifying six behaviors: Feeding, Lying, Standing, Lying-standing-transition, Normal-walking, and Active-walking. The results were validated with public acceleration data. The performance of the proposed classifier has been compared with existing models to highlight the research advantages.