A process for identifying predictive correlation patterns in service management systems

By using the remote functions of a modern IT service management system infrastructure, it is possible to analyze huge amounts of log file data from complex technical equipment. This enables a service provider to predict failures of connected equipment before they happen. The problem most providers face in this context is finding "a needle in a haystack" - the obtained amount of data turns out to be too large to be analyzed manually. The following report describes a process to find suitable predictive patterns in log files for the detection of upcoming critical situations. The identification process may serve as a hands-on guide. It describes how to connect statistical means, data mining algorithms and expert domain knowledge in the domain of service management. The process was developed in a research project which is currently being carried out within the Siemens Healthcare service organization. The project deals with two main aspects: first, the identification of predictive patterns in existing service data and second, the architecture of an autonomous agent which is able to correlate such patterns. This paper summarizes the results of the first project challenge. The identification process was tested successfully in a proof of concept for several Siemens Healthcare products.