LSTM guided homomorphic encryption for threat-resistant IoT networks

Area of Science:

• Cybersecurity
• Network Security
• Data Privacy

Background:

• The Internet of Things (IoT) faces significant security and privacy challenges due to its distributed nature and resource constraints.
• Existing solutions like traditional cryptography and machine learning lack real-time, integrated data privacy and threat resilience.
• Homomorphic Encryption (HE) is computationally expensive, and Long Short-Term Memory (LSTM) networks predict anomalies rather than encrypting data.

Purpose of the Study:

• To propose NeuroCrypt, a novel hybrid system for real-time, privacy-preserving threat detection in IoT networks.
• To address the limitations of existing methods by integrating advanced encryption, anomaly detection, and security management techniques.

Main Methods:

• NeuroCrypt combines Fully Homomorphic Encryption (FHE) with LSTM-based encrypted anomaly detection.
• It incorporates blockchain for dynamic key management and multi-factor authentication.
• The architecture is optimized for edge and fog computing using techniques like ciphertext packing, model quantization, and parallelized encrypted operations.

Main Results:

• The proposed NeuroCrypt framework achieved 99.2% accuracy on a real dataset.
• Performance was evaluated against existing methods including HE-based Deep Neural Networks (DNN), Federated Learning (FL) models, and LSTM Intrusion Detection Systems (IDS).

Conclusions:

• NeuroCrypt offers a privacy-preserving, effective, and scalable solution for real-time threat abatement in IoT environments.
• The hybrid approach successfully integrates encryption, anomaly detection, and robust security management for enhanced IoT security.