Comparing clustering methods for mobile network root cause detection
Foucault, Thomas Etienne (2020)
Foucault, Thomas Etienne
2020
Degree Programme in Information Technology, MSc (Tech)
Informaatioteknologian ja viestinnän tiedekunta - Faculty of Information Technology and Communication Sciences
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2020-05-22
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202004294345
https://urn.fi/URN:NBN:fi:tuni-202004294345
Tiivistelmä
Mobile networks represent a considerable industry globally and are known to rely on robust and highly reliable systems. As a consequence, faults in the system may induce a significant loss of credibility and revenues for mobile operators. However, while mobile network systems implement more features, they also become more complex and difficult to troubleshoot. In response to this issue, this thesis explores the capabilities of cluster analysis methods in order to facilitate the tasks of troubleshooting experts and reduce the cost of mobile networks maintenance.
A comparison of eight different clustering methods is proposed. Each of them is a combination of a dimensionality reduction algorithm (Principal Component Analysis or Self-Organizing Maps) and a clustering algorithm (K-means, OPTICS, or Growing Neural Gas with Post-Pruning), with two exceptions which do not use dimensionality reduction.
The results show that OPTICS performs poorly for this task most of the time. However, K-means and Growing Neural Gas with Post-Pruning demonstrate interesting capabilities for detecting several mobile network faults. Both methods present advantages and disadvantages.
A comparison of eight different clustering methods is proposed. Each of them is a combination of a dimensionality reduction algorithm (Principal Component Analysis or Self-Organizing Maps) and a clustering algorithm (K-means, OPTICS, or Growing Neural Gas with Post-Pruning), with two exceptions which do not use dimensionality reduction.
The results show that OPTICS performs poorly for this task most of the time. However, K-means and Growing Neural Gas with Post-Pruning demonstrate interesting capabilities for detecting several mobile network faults. Both methods present advantages and disadvantages.