A Hybrid Deep-Learning based Security Enhancement Framework for Intrusion Detection in WSNs

Aims and background: Cybersecurity issues have grown as technology rapidly extends the digital landscape. Intrusion-Detection Systems (IDSs) are essential for detecting malicious network traffic. Hardware or software can power these systems. Traditional IDS approaches generally fail to protect data privacy and detect complicated, unique intrusions, principally in WSNs. Objectives and Methods: A hybrid model for WSN intrusion detection utilizing the Convolutional Neural Network and Bidirectional Long-Short-Term-Memory (CNN-BiLSTM) model to overcome these constraints is proposed. Federated-Learning(FL) improves intrusion recognition and privacy in this approach. The FL-based CNN-BiLSTM model is unusual in that numerous sensor nodes can train a central model without revealing private data, addressing privacy issues. By carefully studying local and temporal network links, the CNN-BiLSTM model detects sophisticated and undiscovered cyber threats using deep learning. The WSN-DS and CIC-IDS2017 datasets were used to create the model to detect and classify DoS attacks. Results: Experimental results showed that the FL-CNN-BiLSTM model outperformed existing IDS models in detecting complex and unknown assaults. On both datasets, the model had 99.9% precision and recall, decreasing false positives and negatives. FL and deep-learning (DL) can improve WSN security and privacy, according to our research. Conclusion: The FL-CNN-BiLSTM architecture helps identify complex cyber threats and shows how deep learning may improve intrusion detection systems while protecting user data.