هليا شمس جي

Nowadays the critical cyber-physical systems are more vulnerable to various cyber threats, among which false data injection attacks are particularly insidious. These attacks compromise the integrity an‎d reliability of data by introducing false information, leading to potentially catastrophic consequences by even a small change in data. This thesis investigates the application of Extended Isolation Forest machine learning algorithm to detect these attacks. This algorithm is an enhancement of the stan‎dard Isolation Forest, which improves the performance of the algorithm by introducing ran‎dom slopes an‎d intercepts for branching cuts, making it more effective in high-dimensional data scenarios. To further enhance the detection capability, the Isolation-Based Feature selec‎tion approach was employed. This approach uses the principles of the Isolation Forest to selec‎t features that best separate the normal an‎d anomalous data. By calculating the imbalance score an‎d applying an imbalance penalty, the Isolation Based Feature selec‎tion method identifies the most relevant features that improve the anomaly detection process. The integration of Extended Isolation Forest with this approach ensures that the selec‎ted features not only provide good separability but also enhances the modelʹs performance in identifying attacks. Furthermore data augmentation techniques were utilized to enhance the robustness an‎d precision of the detection model. These techniques involved creating synthetic data points that mimic the distribution of normal data, thereby providing a more comprehensive training set for the machine learning model. The augmented data helps the model better distinguish between normal an‎d anomalous data points, improving its detection capabilities. At last the eva‎luation results demonstrated that the proposed approach could accurately an‎d efficiently detect false data injection attacks. The integration of data augmentation techniques further enhanced the modelʹs robustness, reducing the risk of false positives an‎d false negatives. This research contributes to the field of cybersecurity by providing a novel an‎d effective method for detecting false data injection attacks, thereby strengthening the defenses of critical systems against such threats.

جنگل انزواي تعميم يافته , داده‌افزايي , انتخاب ويژگي مبتني بر انزوا , تزريق داده نادرست

Extended Isolation Forest , Data Augmentation , Isolation Based Feature selec‎tion , False Data Injection

Helia ShamsJey

Mohammadreza Jahed Motlagh