SCADA Intrusion Detection System using Cost Sensitive Machine Learning and SMOTE-SVM
Ms. P. Mahalakshmi, Ramkumar M. P, G. S. R. Emil Selvan · 2022
Emerging technologies throughout the world work with the baseline of Cyber Physical Systems (CPS) to provide remote accessibility. Large scale industries make use of CPS infrastructure, known as Industrial Control Systems (ICS). The ICS using the Supervisory Control And Data Acquisition systems (SCADA) provides remote accessibility, which are highly prone to cyber-attacks. Here arises the need of Intrusion Detection System (IDS). Any Machine Learning (ML) based IDS heavily depends on the historical data, in that case, the class imbalance problem plays a crucial determinant factor of the IDS performance. Thus, the re-search is to address the adverse effects of the class imbalance problem in ML based IDS using a hybrid approach. To do so, the SCADA dataset of the gas pipeline critical infrastructure, is used. To handle the imbalanced data problem, the Synthetic Minority Oversampling Technique with Support Vector Machine (SMOTE_SVM) is adopted as a data level solution combined with an algorithmic level solution, Cost-Sensitive Machine Learning (CSL) is adopted. Different evaluation metrics are used to evaluate the performance of the ML model with and without data balancing techniques. From the results, the hybrid approach of SMOTE-SVM with CSL is proven to be an efficient method to deal with the effects of the imbalanced dataset. The hybrid approach exhibits the following met-rics on using Logistic Regression tuned as CSL; the highest precision and recall scores with 69% and 66% respectively and a zero False Alarm Rate (FAR).