Gradient Boosting Feature Selection With Machine Learning Classifiers for Intrusion Detection on Power Grids

Intrusion Detection in SCADA Based Power Grids: Recursive Feature Elimination Model With Majority Vote Ensemble Algorithm

“…However, they have worked on a subset of the features, and hence we could not identify the exact number of features used in this paper. We have recently proposed WFI based GBFS model for feature selection and extracted 12% of the most promising features in [15]. Our target was to achieve high execution speed and a better predictive model for real-time SCADA communication.…”

“…In our earlier work [15], we have proposed computationally efficient intrusion detection framework for power grids, which not only improves the computational cost but also provides privacy preservation. In that approach, we have determined the most significant features using a Weighted Feature Importance (WFI) based gradient boosting scoring model [15].…”

“…In our earlier work [15], we have proposed computationally efficient intrusion detection framework for power grids, which not only improves the computational cost but also provides privacy preservation. In that approach, we have determined the most significant features using a Weighted Feature Importance (WFI) based gradient boosting scoring model [15]. Furthermore, we have applied eight tree-based algorithms on multilevel multiple datasets to classify various attacks and normal events to validate the efficiency of derived features [15].…”

“…In that approach, we have determined the most significant features using a Weighted Feature Importance (WFI) based gradient boosting scoring model [15]. Furthermore, we have applied eight tree-based algorithms on multilevel multiple datasets to classify various attacks and normal events to validate the efficiency of derived features [15]. In particular, the most promising features were detected by considering multiple values of number of trees while training the model to apply WFI scoring concept.…”

“…We have replaced the single tree base classifier with a majority vote based ensemble method. The selection of various heterogeneous classifiers is based on our preliminary results [15]. This approach will determine the output label based on the majority of the class labels predicted by all the nine classifiers.…”