Cloud failure prediction based on traditional machine learning and deep learning

Abstract Cloud failure is one of the critical issues since it can cost millions dollars to cloud service providers, in addition loss productivity suffered by industrial users. Fault tolerance management key approach address this issue, and prediction techniques prevent occurrence a failure. One main challenges performing produce highly accurate predictive model. Although some work on models has been proposed, there still lack comprehensive evaluation based different types machine learning algorithms. Therefore, paper, we propose comparison model for job task These are built trained using five traditional algorithms three variants deep We use benchmark dataset, called Google Traces, training testing models. evaluated performance multiple metrics determined their important features, as well measured scalability. Our analysis resulted following findings. Firstly, case prediction, found that Extreme Gradient Boosting produces best where disk space request CPU most features influence prediction. Second, Decision Tree Random Forest priority feature both scalability Logistic Regression scalable compared others.