Automatic discovery of rules for predicting network managementevents

摘要

In order to discover behavior patterns, current algorithms onlynanalyze historical data in terms of performance data or fault events,nignoring the temporal correlation among different types of information,nincluding the configuration changes. A method is presented that canndiscover recurrent patterns from multiple flows of events, such asnalarms and configuration events, as well as discrete information, suchnas traffic and usage, taking into account static and dynamic informationnconcerning observed objects and their environments. This method cannfilter out theoretically useless patterns, using a novel technique forndetecting chaos in sequences of events. The prediction accuracy of thendiscovered patterns has been measured using objects with dynamicnbehavior controlled by known and complex differential equations. Thenproposed mining method has been used for discovering and predictingnalarms in a computer network composed of several Internet servers takingninto account the alarm and configuration events history, as well asnstatic information about these servers