Statistical Approach with Machine Learning- Based Intrusion Detection System for CyberAttack Discrimination in the Smart Grid

With rapid power grid digitalisation, keeping the private communications network utilities separate from the public communications networks is increasingly more challenging. It paves the way for the attacker to intrude into the industrial control system by compromising the networks. The proposed framework of Statistical Approach with a Machine Learning classifier (SAML) with Synthetic Minority Oversampling Technique (SMOTE) aims to improve early cyberattack discrimination in the smart grid with optimal hyperparameterized tuning of Principal Component Analysis (PCA) with ExtraTrees and AdaBoost Classifier for Feature Extraction (Dimensionality Reduction), bagging, and boosting, respectively. The significance of the SAML-PCA is that it can handle missing rates by replacing INFinity seen attack records with Zero for the specific column of apparent impedance of the relay to avoid blackouts and cascading failures. The proposed SAML-PCA model achieves a higher accuracy of 95.28% for ExtraTrees with Adaboost Classifier than the ML Classifiers and existing approaches.