Privacy-Preserving Synthetic Data Generation Method for IoT-Sensor Network IDS Using CTGAN.

Area of Science:

• Cybersecurity
• Data Privacy
• Machine Learning

Background:

• Internet of Things (IoT) networks face increasing privacy risks due to large datasets used in machine learning (ML) for intrusion detection systems (IDS).
• Reliance on third parties for data storage and ML model training exacerbates these privacy concerns.
• Existing methods struggle to balance data utility for IDS with robust privacy preservation.

Purpose of the Study:

• To propose a novel privacy-preserving synthetic data generation method for IoT sensor network data.
• To maintain the utility of data for ML-based IDS while safeguarding sensitive information.
• To mitigate privacy risks associated with data sharing and third-party processing.

Main Methods:

• Utilized a conditional tabular generative adversarial network (CTGAN) for synthetic data generation.
• Integrated differential privacy (DP) with CTGAN through controlled noise injection.
• Employed dynamic distribution adjustment and quantile matching to optimize the utility-privacy tradeoff.

Main Results:

• Achieved a significant improvement in data utility compared to standard DP methods, evidenced by a KS test score of 0.80.
• Effectively minimized privacy risks, including singling out, linkability, and inference attacks.
• Demonstrated the capability of synthetic datasets to support effective intrusion detection.

Conclusions:

• The proposed DP-enhanced CTGAN method offers a robust solution for privacy-preserving synthetic data generation in IoT environments.
• This approach successfully balances data utility for IDS with strong privacy guarantees.
• Enables secure utilization of IoT data for developing effective intrusion detection systems without compromising user privacy.