Sheep's coping style can be identified by unsupervised machine learning from unlabeled data

Area of Science:

• Animal Behavior
• Machine Learning
• Animal Science

Background:

• Understanding animal behavior is crucial for welfare and management.
• Coping styles, or behavioral responses to stress, vary among individuals.
• Objective methods are needed to classify these styles in livestock.

Purpose of the Study:

• To define and classify sheep coping styles using unsupervised machine learning.
• To identify distinct behavioral patterns associated with different coping strategies in sheep.
• To validate the use of machine learning for objective behavioral classification in animal science.

Main Methods:

• Applied Principal Components Analysis (PCA) to behavioral data from 105 Norduz sheep in an arena test.
• Utilized Agglomerative Hierarchical Clustering (HCA) on PCA components to identify sheep coping styles.
• Compared multiple clustering methods (k-means, various hierarchical linkages) and validated cluster stability using bootstrap resampling and Jaccard coefficients.

Main Results:

• Identified three distinct sheep coping styles (CS): reactive (n=71), intermediate (n=22), and proactive (n=12) using Ward's method with 3 clusters.
• Coping style significantly affected behavioral variables including distance to group (DTG), distance to stimulus (DTS), and locomotion (LOC).
• Proactive sheep exhibited higher activity levels (DTG, DTS, LOC, SCR) compared to reactive sheep, while intermediate sheep showed intermediate behavioral patterns.

Conclusions:

• Distinct coping styles in sheep can be objectively identified through behavioral analysis in arena tests.
• Unsupervised machine learning provides a robust framework for classifying sheep coping styles from unlabeled behavioral data.
• These findings contribute to a deeper understanding of individual behavioral variation and its implications for sheep management and welfare.