Enhanced Grid Stability Assessment Using Cost-Sensitive Ensemble Learning and AI-Driven Instability Prediction for Decentralized Smart Grids

Among the most important aspects of smart city sustainability involves centralized smart grids, which are the subject of this research. Smart grids use sophisticated technology to effectively handle energy, combining both clean and dirty sources. Stability is greatly affected, however, by problems including power fluctuations, theft, and data administration. A new Cost-Sensitive Stacked Ensemble Classifier (CS-SEC) is suggested to solve these problems; it combines a Support Vector Machine meta-classifier with four machine learning algorithms: XGBoost, Naive Bayes, Random Forest, and Nu-Support Vector Machine (Nu-SVM). By selecting features based on correlation coefficients, the model improves the speed of computing and decreases redundancy. In terms of performance, the CS-SEC model is head and shoulders above the competition, with an AUC of 0.99, sensitivity of 99%, specificity of 98.34%, and accuracy of 98.57%. The ensemble method is more effective in improving predicting accuracy and decreasing misclassification when compared to solo classifiers. Optimal and stable smart grid operations may be achieved with the help of sophisticated ensemble techniques, according to the results. This opens the door to smart city infrastructures that manage electricity reliably and sustainably.