Monitoring and analysis of animal behavior are two of the prominent applications of Wireless Sensor Networks (WSN) in modern Dairy Farming. Behavioral information collected by sensor devices worn by the animals is expected to provide early detection of stress and onset of specific diseases. Animal mobility coupled with farm-based contextual information is expected to automate and increase efficiency of the pasture. Though some WSN solutions have been proposed for these applications, their realizations commonly depend on high availability of third-party components (e.g. cloud-environment for behavior analysis). This reduces suitability of these solutions for pasture-based dairy farms, where large scale and remote locations significantly restrict accessibility to external components (e.g. poor or no internet connectivity). Meanwhile, continuous design improvement of WSN devices has significantly increased their computational capacity. To take advantage of this, a novel Edge Mining (EM) concept has been proposed under the umbrella of Fog Computing, where to increase availability, data analysis is partially hosted by WSN. In this article, we propose an Edge Mining implementation of our WSN system for analyzing animal mobility and behavior. We develop a novel EM method that could be used for a range of animal activity and behavior analysis. Performance of the method is evaluated regarding the accuracy and suitability for WSN-based execution.