DeepSense: An Adaptive Scalable Ensemble Framework for Industrial IoT Anomaly Detection

Area of Science:

• Cyber-Physical Systems Security
• Industrial Automation and Control
• Machine Learning for Security

Background:

• Industrial Internet of Things (IIoT) environments face increasing security challenges due to scale, heterogeneity, and dynamic behavior.
• Conventional security mechanisms struggle to cope with the expanded attack surface in modern IIoT deployments.
• Resource-constrained and heterogeneous IIoT systems require specialized security solutions.

Purpose of the Study:

• To propose DeepSense, a hybrid and adaptive anomaly and intrusion detection framework for IIoT.
• To develop a realistic data pipeline and experimental testbed (DataSense) for synchronized sensor and network data processing.
• To create a comprehensive evaluation framework assessing detection quality, latency, resource efficiency, and coverage.

Main Methods:

• DeepSense integrates three components: DataSense (data pipeline/testbed), RuleSense (edge-based rule detection), and NeuroSense (adaptive ML/DL ensemble).
• NeuroSense utilizes an ensemble of 22 models (classical, neural, hybrid, Transformer) for validating suspicious events and classifying attacks.
• A Pareto-optimal ensemble selection process is employed under realistic IIoT constraints.

Main Results:

• DeepSense demonstrated strong generalization capabilities across diverse detection scenarios.
• The framework achieved lower false positive rates and robust performance against evolving attack behaviors.
• Experimental results validated the scalability and efficiency of DeepSense for IIoT security.

Conclusions:

• DeepSense provides a scalable and efficient security solution tailored for resource-constrained and heterogeneous IIoT environments.
• The framework meets the operational demands of Industry 4.0 and the resilience goals of Industry 5.0.
• DeepSense offers a promising approach to enhance the security posture of industrial cyber-physical systems.