AI-Driven Predictive Maintenance and Fault Detection for Cloud Infrastructure Using the Stacking Ensemble Method

Abstract:

The cloud infrastructure must be very reliable because their sudden downtime creates a high economic cost to the clients. Therefore, the following project presents an artificial intelligence-driven predictive maintenance system featuring a fault detection system in the cloud environment. It applies the stacking ensemble method as an approach toward enhancing the efficiency of its detection model. The act of fusing the outputs of multiple machine learning algorithms with every model leads effectively to good predictions of imminent failure and hence allows proactive maintenance to be scheduled. four datasets were trained with the following models Random Forest, gradient boosting, AdaBoost model, Support Vector Machine, K-nearest Neighbours, Logistic regression, Artificial Neural Network, XGBoost, and Stacking Classifier, these model performance was evaluated to show how well they performed when predicting maintenance by detecting a fault in cloud-based performance metrics used are Accuracy, Precision, Recall, Specificity, MCC, Kappa, F-Score, AUC, FDR, FNR, FPR, NPV, across the four datasets the stacking model performed the best for the first analysis we obtained the accuracy of 0.9830, second dataset accuracy of 0.950249, third analysis the accuracy score is 1.0000, and the fourth as 0.962442 Results show that stacking ensemble methods have much better predictive performance compared to single-model-based methods, and the approach is efficient in overcoming disruptions in the cloud infrastructures

Keywords: Cloud environment, predictive maintenance, fault detection, stacking ensemble method