An Algorithm for Clustering Animals by Species based upon Daily Movement

An algorithm is presented for clustering individual animals by species based solely upon the daily movements of the individual animals. This is particularly challenging due to the highly erratic nature of the animals’ movement. The variance in the scale and frequency of movement between individuals within a species is often greater than the difference between species. Existing clustering algorithms including hierarchical, k-means, and spectral were tried, but they failed to accurately distinguish between species or to cluster individuals of the same species together. Also, some of these algorithms require a priori knowledge of the number of species (clusters). The algorithm presented here addresses this problem by creating separation through a distance metric based upon ranking and then clustering based upon commonality in rankings. The algorithm performed well, demonstrating the ability to both distinguish between species and to cluster together individual of the same species. Furthermore, it assumes no a priori knowledge of the number of clusters expected.