Proceedings of the
The Nineteenth International Conference on Computational Intelligence and Security (CIS 2023)
December 1 – 4, 2023, Haikou, China
A Multi-Meta Stacking Ensemble Learning Approach for Intrusion Detection in Industrial Control Systems
1School of Cyber Engineering, Xidian University, China.
2Hi-Think Technology, Corp, China
ABSTRACT
With the rapid advancements in Internet technology and industrial control systems (ICS), the security of ICS is facing an increasing number of threats. The complex structure of ICS and the proliferation of attack types have resulted in low accuracy and high false positive rates in current intrusion detection techniques. To address these issues, we propose a novel approach for ICS intrusion detection called multi-meta stacking ensemble learning. This approach involves training two meta-classifiers using the outputs of multiple diverse base classifiers. The final decision is made by combining the classifications from the two meta-classifiers through a soft voting mechanism. Experimental results on the well-known gas pipeline dataset demonstrate the exceptional performance of our proposed method, achieving an impressive accuracy of 99.65%, a low false positive rate of 0.86%, and an extremely low false negative rate of 0.52%. When compared to other intrusion detection models, our proposed method exhibits superior detection performance.
Keywords: Intrusion detection, Industrial control system.
Download PDF
1School of Cyber Engineering, Xidian University, China.
2Hi-Think Technology, Corp, China
ABSTRACT
With the rapid advancements in Internet technology and industrial control systems (ICS), the security of ICS is facing an increasing number of threats. The complex structure of ICS and the proliferation of attack types have resulted in low accuracy and high false positive rates in current intrusion detection techniques. To address these issues, we propose a novel approach for ICS intrusion detection called multi-meta stacking ensemble learning. This approach involves training two meta-classifiers using the outputs of multiple diverse base classifiers. The final decision is made by combining the classifications from the two meta-classifiers through a soft voting mechanism. Experimental results on the well-known gas pipeline dataset demonstrate the exceptional performance of our proposed method, achieving an impressive accuracy of 99.65%, a low false positive rate of 0.86%, and an extremely low false negative rate of 0.52%. When compared to other intrusion detection models, our proposed method exhibits superior detection performance.
Keywords: Intrusion detection, Industrial control system.
Download PDF