Federated Edge Intelligence for Privacy Preserving Livestock Disease Prediction in Smart Farming Environments

Abstract

The rising adoption of smart farming technologies has enabled new approaches to predict early stages of livestock diseases through continuous monitoring of physiological and behavioural parameters. But conventional cloud centric approaches raise serious concerns about data privacy, communication overhead and scalability for geographically distributed farms. To overcome these challenges, this paper presents a Adaptive Federated Edge Intelligence (AFEI) framework for privacy preserving livestock disease prediction in smart farming environments. The proposed framework combines IoT enabled livestock sensing, edge intelligence, federated learning, differential privacy and a Transformer BiLSTM disease prediction model in a single architecture. The Edge Disease Risk Score (EDRS) mechanism is used to prioritise disease sensitive samples. The Adaptive Disease Aware Aggregation (ADAA) strategy is employed to improve the learning of the global model by dynamically weighting the participating clients based on the quality of the local models and the reliability of the data. Experimental results show that the proposed framework can achieve better disease prediction performance with an accuracy of 95.8% and reduce the communication overhead by 40.4% compared with the conventional federated learning approaches. The results show the efficiency of the proposed framework in terms of secure, scalable and communication efficient livestock health monitoring.