Applications and trends of accelerometer and artificial intelligence techniques in livestock production: a scoping review

• Deep neural networks are trending in livestock accelerometer data analysis. • Larger datasets and custom devices need further development. • Energy efficiency and cost are key for large-scale adoption. • The integration of multiple sensors and real-time classification requires further exploration. • The combination of accelerometers with computer vision and physiological sensors enhances precision livestock farming. This study presents a scoping review of the use of accelerometers combined with artificial intelligence (AI) techniques in livestock production, focusing on applications, methodological characteristics, and emerging trends. Following PRISMA guidelines, 108 peer-reviewed studies were analyzed, revealing a marked increase in publications from 2021 onward. The review addressed advances in sensor technologies and AI methods, target species, monitored variables, experimental scale, sensor configuration, data processing strategies, and modeling approaches. Most studies focused on cattle (68.52%), followed by sheep (20.37%), pigs (5.56%), poultry (3.70%), and goats (1.85%). Applications were predominantly related to behavior classification, estrus and parturition detection, disease monitoring, and feed intake or nutritional assessment. A recurrent limitation across the literature was the small experimental scale, with approximately 50% of studies involving 18 animals or fewer, potentially constraining model generalization. In addition to accelerometers, complementary sensors such as GNSS, gyroscopes, and temperature sensors were frequently integrated to enhance behavioral and contextual inference. Data processing workflows commonly included signal cleaning, filtering, normalization, segmentation, and feature extraction in time and frequency domains. Modeling approaches often incorporated data augmentation, class balancing, cross-validation, and hyperparameter tuning. Traditional machine learning models (e.g., Random Forest, SVM, k-NN, and Decision Tree) remain prevalent, while deep learning architectures, particularly CNNs and LSTMs, have gained increasing attention in recent years. Overall, current trends suggest growing interest in multimodal sensing, customized devices, and real-time classification, alongside persistent challenges related to energy efficiency, economic feasibility, and scalability for commercial livestock systems.