Optimizing heat stress detection in dairy cattle: leveraging datamining and unsupervised analyses to explore individual-level impact through behavior and meteorological factors

Dairy cows have various strategies for coping with heat stress, including a change in behavior. The aim of this study was to explore, via unsupervised learning, the behavior of cows and to relate it to their environment using the comprehensive climate index. A total of 8,928 observations, associated with behaviors known to be influenced by heat stress, per cow over the month of August 2020 were recorded from a herd of 28 grazing cows. The CCI was established for each day using radiation, relative humidity, ambient temperature and wind speed. Hopkins statistic was used to measure the clustering potential of the observations, with a value of 0.812 indicating strong clustering of the data. A principal component analysis was performed to determine the number of groups to be formed
based on the data. Visualization of dimensions 1 and 2, which explain 58.81% of the
ariability in
the data. The unsupervised learning method of k-means partitioning was implemented in order to form 4 distinct groups and outliers in each group were removed using the Mahalanobis distance method based on a p-value of less than 0.05. The interpretation of the groups was based on the average of the behaviors. A correlation of 0.44 was established between the first group and the increase of CCI. The potential prospects of this study are to provide a better understanding of the individual responses of cattle to heat stress and to improve health management. In addition to an approach based on behavior and not on an index, future predictive models could subsequently be implemented to enable early adaptation in the face of events unfavorable to animal welfare.