PreCog : a robust machine learning system to predict failure in a virtualized environment

Abstract

The research in this work addresses the need for a warning system to predict future application failures. PreCog, the predictive and regressional error correlating guide system, aims to aid administrators by providing a robust future failure warning system statistically induced from past system behavior. In this work, we show that with the use of machine learning techniques such as Adaptive Boosting and Correlation-based Feature Selection, PreCog, without any prior knowledge of its target, can be accurately and reliably trained within a virtual environment using past system metrics to predict future application in a variety of domains.