Privacy-Preserving Detection of Tampered Radio-Frequency Transmissions Utilizing Federated Learning in LoRa Networks

Area of Science:

• Cybersecurity
• Internet of Things (IoT)
• Machine Learning

Background:

• LoRa networks are crucial for low-power, long-range IoT communication.
• Radio-frequency transmissions in LoRa are vulnerable to tampering and attacks.
• Existing security methods may compromise data privacy.

Purpose of the Study:

• To develop a privacy-preserving method for detecting tampered transmissions in LoRa networks.
• To identify unknown attacks in LoRa-based IoT systems.
• To leverage Federated Learning (FL) for secure anomaly detection without sharing raw data.

Main Methods:

• Evaluated multiple FL-enabled anomaly detection algorithms: CAE-FL, IF-FL, OCSVM-FL, LOF-FL, and K-Means-FL.
• Trained models on distributed devices using FL to maintain data privacy.
• Assessed performance using accuracy, precision, recall, and F1-score.

Main Results:

• Convolutional Autoencoder Federated Learning (CAE-FL) demonstrated superior performance with 97.27% accuracy and 0.97 for precision, recall, and F1-score.
• Isolation Forest Federated Learning (IF-FL) achieved 96.84% accuracy.
• Clustering-based methods like OCSVM-FL, LOF-FL, and K-Means-FL showed robust detection capabilities.

Conclusions:

• Federated Learning effectively enhances privacy and security for anomaly detection in LoRa networks.
• The proposed FL approach can detect tampered signals and unknown attacks, securing sensitive IoT applications.
• CAE-FL is a highly effective algorithm for privacy-preserving anomaly detection in LoRa communications.