Abstract
Accurate failure predictions can help in mitigating the impact of computer failures. Resources, applications, and services can be scheduled to limit the impact of failures. However, providing accurate predictions with sufficient lead of time is challenging. Log files track changes of system state. A sequence or a pattern of messages may be used to predict failures. Here we describe an approach to predict failures based on the Cox Proportional Hazards (PH) model that has been applied successfully in various fields of research. We apply our approach to log files collected during approximately 3 months of work in a Urge Italian company. We compare the performance of the proposed model with Support Vector Machines.