Table 2.

The results of AlphaTracker's clustering algorithm are consistent with human judgement.

	Large dataset (3,034 clips)	Small dataset (1,345 clips)	Random assignment
Adjusted rand index	0.201186	0.186533	0.003451

The accuracy of AlphaTracker's clustering algorithm was evaluated by comparing its output with human annotations for individual clips (500 ms). The annotations included individual behaviors: walking, digging, sniffing, rearing, turning, face grooming, and body grooming, and nine social behaviors: following, chasing, anogenital sniffing, face sniffing, and social rearing. The similarity of the class assignments was measured using the Adjusted Rand Index (ARI). AlphaTracker showed significantly higher ARI compared to random assignment, demonstrating its consistency with human judgement. Additionally, using a relatively small dataset of 1,345 clips, AlphaTracker was able to accurately capture most of the cluster assignments, further highlighting its efficiency and effectiveness.