Enhanced reliability in wireless sensor networks: federated learning with Bayesian weighted random forest for secure and privacy-preserving threat detection

Intrusion Detection Systems (IDS) are essential for identifying anomalous network traffic that may indicate malicious activity, particularly with the rapid growth of technology and escalating cybersecurity threats. However, traditional IDS solutions both hardware and software often struggle to ensure data privacy and accurately detect sophisticated or previously unseen attacks, especially in Wireless Sensor Networks (WSNs). To overcome these challenges, this paper proposes an Enhanced Reliability Model for Wireless Sensor Networks: Federated Learning with Bayesian Weighted Random Forest (WSN-FL-BWRF) for secure and privacy-preserving threat detection. The proposed method leverages Federated Learning (FL) to enable multiple sensor nodes to collaboratively train a global intrusion detection model without sharing sensitive local data. The Bayesian Weighted Random Forest component enhances the detection of complex and emerging threats by analyzing both local and temporal network patterns. The WSN-FL-BWRF model is specifically developed to detect and classify various Denial of Service (DoS) attacks using WSN-DS dataset. Experimental evaluations demonstrate that the proposed method outperforms conventional Artificial Deep Neural Network (ADNN) models, with high precision and recall and reduced false positive and negative rates. Compared to other federated learning models, such as WSN-SCNN-Bi-LSTM, WSN-Ensemble FL, and WSN-FDRL, the proposed model exhibits superior detection accuracy. While the model achieved performance metrics nearing 99.9% under certain conditions, further validation on diverse datasets is necessary to assess generalizability and mitigate risks of overfitting or dataset-specific bias. This research underscores the potential of combining federated learning with advanced ensemble techniques to enhance both the safety and privacy of WSNs in real-world applications.