This study investigates the use of machine learning to classify contingency scenarios in power systems into safe, moderate and severe levels, aiming to enable proactive decisions and reduce large‑scale outage risks. System states are obtained via Newton‑Raphson load‑flow analysis, and the Overall Performance Index (OPI) serves as the safety metric. To address class imbalance and high dimensionality, Synthetic Minority Over‑Sampling Technique (SMOTE) and Principal Component Analysis (PCA) are applied, yielding four preprocessing configurations: normalized, SMOTE, PCA, and SMOTE+PCA. Datasets are generated from N‑k (k=1,2,3) contingency cases on IEEE‑14 and IEEE‑30 bus systems, and three classifiers—K‑Nearest Neighbours (KNN), Random Forest (RF) and Support Vector Machine (SVM)—are trained and evaluated. Performance is measured by precision, recall and F1 score, with emphasis on the severe class. Results show RF achieves the highest F1 (0.97 on IEEE‑30, 0.86 on IEEE‑14), SVM benefits markedly from PCA, and KNN performs best with the SMOTE+PCA combination. Overall, PCA contributes more to model performance than SMOTE; while SMOTE improves recall, it may introduce false positives and reduce accuracy. The findings demonstrate that machine‑learning‑based contingency screening offers a scalable, real‑time alternative to traditional analysis, enhancing power system security assessment.
Review: The work provides a thorough comparison of three popular classifiers and two data‑balancing/dimensionality‑reduction techniques, highlighting the pivotal role of feature reduction in boosting model robustness and offering a practical ML framework for power‑system security monitoring.