RQGPR: Rational Quadratic Gaussian Process Regression for Attack Detection in the SCADA Networks

Abstract

The constant development and deployment of the supervisory control and data acquisition (SCADA) in the industrial internet of things (IIoT) have enabled vast communication leading to the generation of large volumes of sensor data. This phenomenon has increased SCADA's susceptibility to vulnerability and attacks which calls for attack detection mechanisms. Existing systems only aim at detection accuracy without considering the effect of false alarm rates in large sensor data. To resolve this issue, we propose a Rational Quadratic Gaussian Process Regression (RQGPR) for the effective reduction of false alarm rate and improved prediction precision. In this algorithm, a Gaussian process regression model is trained with recourse to kernel functions to precisely predict attacks and reduce false alarms. The RQGPR outperforms all other kernels in the reduction of false alarm rates. Through simulations, we show that the proposed model reduces the false alarm rate up to 71.73% higher than other kernels. This result was validated by evaluating the CIRACIC-DoHBrw-2020 datasets, which also had a reduction rate of 67.61%. In addition, it also showed superior performance when compared with other state-of-the-art models.