A Machine Learning Framework for Predicting Failures in Cloud Data Centers -A case of Google Cluster -Azure Clouds and Alibaba Clouds

Résumé

Abstract The large scale and dynamic nature of cloud has added extra complexity when it comes to fault detection and management. Availability directly depends upon how fast the cloud infrastructure can detect any faults and take necessary steps to troubleshoot the problem. It is critical for service providers to provide stable service or else it may cause losses for clients. It is important to detect a failure in its embryonic stages to employ preventive measures to avoid a disastrous failure before it occurs. Researchers have focused on pure-bred failure characterization and analysis machine learning models to enhance cloud failure prediction accuracy for large cloud data centers but limited research has been done with ensemble models. In this paper we develop an enhanced cloud failure prediction model based on Adaboost ensemble Machine Learning algorithms that predicts hardware and software failures using Google Cluster 2019, Azure Clouds and Alibaba clouds datasets. Our model employs ensemble classification using Logistic Regression, Random Forest Classifier and Decision Tree Classifier. Results indicate our approach recorded a marginal improvement in accuracy prediction compared with results from previous researchers in the area. Decision Tree yielded best average model performance results recording 91.7% Precision, 88.8% Recall, 89.7% F1 Score, 94.0% Accuracy and 89.0% AUC.

Citer ce document

Accès au document

Texte intégral en lecture en ligne, réservé aux abonnés SPHAERO et aux membres de l'institution. Se connecter

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0