Anomaly Detection in Industrial IoT Using Distributional Reinforcement Learning and Generative Adversarial Networks

Area of Science:

• Cybersecurity
• Artificial Intelligence
• Industrial Internet of Things (IIoT)

Background:

• Anomaly detection is critical for Industrial Internet of Things (IIoT) security due to increasing cyber threats.
• Intrusion Detection Systems (IDS) are vital for monitoring and protecting IIoT networks and critical infrastructure.
• Data imbalance in IIoT security datasets hinders the effective detection of rare but critical attacks.

Purpose of the Study:

• To propose a novel mechanism for enhancing the efficiency and robustness of IDS in IIoT environments.
• To address the challenge of data imbalance in anomaly detection using artificial data generation.
• To improve the detection rates of minority attacks within IIoT networks.

Main Methods:

• Integration of Distributional Reinforcement Learning (DRL) with Generative Adversarial Networks (GAN) for IDS.
• Utilizing GAN to generate realistic and balanced feature distributions from artificial data.
• Evaluating the DRL-GAN model on the Distributed Smart Space Orchestration System (DS2OS) dataset for anomaly detection.

Main Results:

• The DRL-GAN model demonstrated superior performance compared to the standard DRL model in anomaly detection.
• Significant improvements were observed in accuracy, precision, recall, and F1 score for both binary and multiclass classifications.
• The GAN's role in enhancing DRL training proved effective in improving the detection of specific, underrepresented data classes (minority attacks).

Conclusions:

• The proposed DRL-GAN approach offers a robust solution for enhancing IDS in IIoT environments.
• Generative Adversarial Networks effectively mitigate data imbalance issues, boosting the detection of low-frequency cyber threats.
• This hybrid AI model represents a significant advancement in securing critical IIoT infrastructure against sophisticated cyber attacks.