A systematic review of artificial intelligence in small ruminant production systems: applications, performance outcomes, and reported implementation challenges.

Abstract

Background: Sheep and goats are essential components of global livestock systems, supporting smallholder livelihoods and contributing substantially to meat, milk, and fiber production. However, many small ruminant production systems remain extensive and constrained by limited infrastructure, labor availability, and challenges in continuous animal monitoring. Artificial Intelligence (AI) offers new opportunities for precision livestock management through automated data analysis, early detection of production- and health-related changes, and improved decision support. This systematic review aimed to characterize current AI applications in sheep and goat production, summarize reported performance outcomes across all major application domains, and identify barriers affecting practical implementation.

Results: A systematic search of Web of Science, Scopus, and PubMed identified peer-reviewed studies published between January 2020 and December 2025. From 11,035 records screened, 92 studies met the inclusion criteria and were synthesized narratively; meta-analysis was not conducted due to substantial methodological heterogeneity across studies. AI applications spanned six domains: behavior and activity recognition (26.1%, n = 24; mean accuracy 92.4%, range 66.7-100%), individual animal identification (19.6%, n = 18; mean accuracy 97.3%, range 93.3-99.9%), health, welfare, and disease detection (19.6%, n = 18; mean accuracy 89.7%, range 62.0-99.0%), growth and body measurement (9.8%, n = 9; mean R² = 0.86), genomics and molecular biology (8.7%, n = 8; mean accuracy 97.8%), and production, technical, and environmental applications (16.3%, n = 15; mean accuracy 93.1%). Convolutional Neural Networks, YOLO-based models, and Random Forest algorithms were the most frequently applied approaches. Publications grew markedly over the review period, with 28.3% published in 2025 alone. Fewer than half of included studies used fully independent external validation. Key implementation barriers included limited dataset diversity, class imbalance, environmental complexity, and hardware constraints.

Conclusions: Current evidence indicates that AI has strong potential to enhance small ruminant production, particularly for automated monitoring and biometric identification under controlled conditions. However, translation into sustainable real-world applications remains limited by methodological inconsistencies, restricted validation across farms and breeds, and insufficient representation of extensive production environments. Future progress will require standardized public datasets, transparent domain-appropriate metric reporting, rigorous independent validation, and deployment-focused evaluation under practical farming conditions.