Characterization of Cattle Behavior Based on Dimensional Reduction Techniques

Accelerated human population growth and global economic development have significantly increased the demand for food, particularly for dairy and meat products, challenging the livestock industry to seek more efficient and sustainable management approaches. Precision Livestock Farming (PLF), focused on individualized cattle monitoring, has positioned itself as a key solution to this challenge. In this sense, activity monitoring collars represent a promising innovation, allowing detailed, real-time observation of the behavior of each animal. In this context, this paper analyzes and compares three dimensional reduction techniques (Kernel PCA, Laplacian Eigenmaps, and UMAP) to characterize and classify the daily behavior of dairy cows in an intensive farm based on information obtained through activity monitoring collars. The results achieved have shown how UMAP stands out as a particularly effective technique as visual tool to individualized or small-group identification of cows with similar patterns. The capacity for characterization and classification is crucial as a preliminary step for developing predictive models focused on detecting anomalous events, such as diseases, calving, or estrus, thus enhancing the efficiency of herd management and contributing to the sector’s sustainability.