Wireless Sensor Networks (WSNs) are essential to next-generation wireless systems, their scattered deployment and limited resources make them extremely susceptible to cyberattacks. While current deep learning-based intrusion detection systems (IDS), such CNN, LSTM, and GRU models, achieve excellent accuracy, they have drawbacks, including high processing costs, centralized training, and privacy concerns. This research suggests a Federated Attention-Based Lightweight Intrusion Detection System (FA-IDS) for WSNs in order to overcome these constraints. In order to enable decentralized training among sensor nodes while maintaining data privacy and lowering communication cost, the suggested approach combines federated learning (FL) with a Lightweight Attention Enhanced BiGRU (Att-BiGRU) classifier. To balance node energy consumption and detection accuracy, an energy-aware loss function is implemented. When compared to centralized CNN, LSTM, and GRU techniques, experimental evaluation on the WSN-DS dataset shows that FA-IDS achieves 99.1% detection accuracy, reduces energy consumption by 18–25%, and dramatically lowers communication cost. The outcomes verify that the suggested method is suitable for scalable and safe WSN deployments .
 

Wireless Sensor Networks, Intrusion Detection System, Federated Learning, Attention Mechanism, Energy Efficiency.