Improving Monitoring of IT Systems using Machine Learning

High availability of server-bound business applications has become crucial in today's IT landscape. It is therefore important to monitor these systems for any deviations. Machine learning can be implemented to improve IT monitoring. In this project, an anomaly detection using neural networks was implemented on multidimensional real world server data. Having a proper anomaly detection enables: faster detection of deviations; better capacity for proactive maintenance; and more rapid root cause analysis. Using machine learning to improve IT monitoring or monitoring of data from other domains in a similar format is neither a new topic in the industry nor academia. Prior work in the industry is often a black box and difficult to evaluate, and previous attempts in the academia that use a similar approach as this project implements run their experiments on different kinds of data. The neural networks were assessed using different configurations. Additionally, the ideal configuration was then compared to an established anomaly detection algorithm. The results look promising, but more data and work is needed to evaluate and tune the approach systematically.

Author dc:creator

• Guðlaugur Garðar Eyþórsson 1991-

Contributors dc:contributor

• Háskólinn í Reykjavík

• Tölvunarfræði
• Gagnanám
• Meistaraprófsritgerðir
• Computer science
• Machine learning

Language dc:language.iso

en

Handle dc:identifier.uri

http://hdl.handle.net/1946/28745

OAI identifier oai:identifier

oai:skemman.is:1946/28745