Classifying and Understanding of Dairy Cattle Health Using Wearable Inertial Sensors With Random Forest and Explainable Artificial Intelligence

Recent developments in the field of machine learning andwearable sensor technology have led to a renewed interest in the efficient monitoring of cattle health conditions. Wearable inertial sensors are particularly suited for a noninvasive solution to assess cattle health by continuously providing their daily behavior information. However, current classification methods have limitations in terms of the lack of interpretability analysis of the resulting models for the combined behaviors. In this study, a data-driven assessment approach was proposed to classify and understand cattle health through behavior recognition and health classification algorithms. Data were collected through wearable sensors attached to the cattle's collar and leg, providing acceleration, angular velocity, and Euler angles. Random forest was used to implement behavior recognition and health classification of dairy cattle, which was incorporated with explainable artificial intelligence (XAI) for result interpretation. The behavior recognition model achieves high accuracy in distinguishing between behaviors. The health classification model exhibits a strong discriminative ability, and the receiver operating characteristic curve confirms its effectiveness in identifying cattle health status using behavior data. The results showcase the crucial role of XAI analysis in establishing a correlational description between the combined behavioral characteristics of cows and their health classification. This work not only enhances the credibility of the development model but also offers valuable guidance for the predictions of cattle health by motion and feeding behaviors.