USING DATA MINING AND MACHINE LEARNING METHODS FOR SERVER OUTAGE DETECTION - MODELLING NORMALITY AND ANOMALIES

摘要

This paper will discuss several approaches to detect abnormal events, which are considered to be worth further investigation by the modeler, in a time series of frequently collected data as early as possible and -wherever applicable - to predict them. The approaches to this task use various methods originating in the field of data mining, machine learning and soft computing in a hybrid manner. After a basic introduction including several areas of application,the paper will focus on the modular parts of the proposed methodology, starting with a discussion about different approaches to predict time series. After the presentation of several algorithms for outlier detection, which are applicable not only for time series, but also a chain of events, the results of the simulation gained in a project to detect server outages as early as possible are put up for discussion. The text ends with an outlook for possible future work.