A Federated Edge Artificial intelligence (Edge-AI) Framework For Livestock Health Monitoring

Precision livestock farming increasingly relies on intelligent monitoring systems to improve animal health, welfare, and farm productivity. However, conventional cloud-centric approaches often face challenges related to data privacy, communication overhead, network dependency, and real-time decision-making. This study proposes a Federated Learning (FL) and Edge-AI framework for distributed livestock health monitoring that integrates Smart Neck Collar sensing, edge computing, MQTT-based communication, and a hybrid Gradient Boosting–Long Short-Term Memory (GBoost-LSTM) prediction architecture. The proposed framework enables local preprocessing, model training, and inference at participating farms while supporting collaborative model learning without transferring raw livestock sensor data to a central server. A simulation-based federated evaluation was designed using livestock health data collected from 150 dairy cows across seven districts of Punjab, India. The proposed framework was compared with previously evaluated centralized machine-learning approaches to assess its predictive performance. The simulated FL-GBoost-LSTM model achieved an accuracy of 95.12%, precision of 93.48%, recall of 88.21%, and F1-score of 90.76% under the evaluation configuration described in this study. The results indicate that federated collaborative learning can provide competitive disease-prediction performance while retaining data locally at participating farms. In addition, the proposed edge-based architecture is designed to reduce dependence on continuous cloud connectivity and support localized health-status assessment. Because the present evaluation is simulation-based and does not constitute a field deployment, the reported communication, latency, privacy, and scalability benefits should be interpreted as framework-level findings rather than validated operational outcomes. Future work will focus on real-world multi-farm deployment, systematic non-IID evaluation, formal privacy-enhancement mechanisms, and comprehensive benchmarking of computational, communication, and energy requirements. Proposed a Federated Learning and Edge-AI framework for distributed livestock health monitoring across geographically separated dairy farms. Integrated Smart Neck Collar sensing, MQTT-based communication, edge computing, and hybrid GBoost-LSTM prediction within a unified architecture. Simulated collaborative model training across distributed farm-level datasets while retaining raw livestock data locally. Achieved 95.12% accuracy, 93.48% precision, 88.21% recall, and 90.76% F1-score under the reported simulation-based evaluation configuration. Compared the proposed federated framework with previously evaluated centralized livestock disease-prediction models to assess predictive performance. Identified communication efficiency, edge inference, non-IID data handling, privacy enhancement, and real-world deployment as key areas requiring further experimental validation.