Bayesian methods and decision trees to predict failuredynamics in cloud computing systems.

. In contrast to classical reliabilitymethods, failure prediction is based on runtime monitor-ing and a variety of models and methods that use thecurrent state of a system and the past experience as well.Most of the existing failure prediction methods arebased on statistical learning techniques [3]. They usesupervised learning models to approximate the depen-dency of failure occurrences on various performancefeatures [1], [4]–[6]. The underlying assumption of thosemethods is that the training dataset is labeled,

10A Survey of Online Failure Prediction Methods

A Closed-Loop Automatic Data-Mining Approachfor Preventive Network Monitoring