%0 Conference Paper %1 fie19ITC31 %A Fiedler, Markus %B 31th International Teletraffic Congress (ITC 31) %C Budapest, Hungary %D 2019 %K itc itc31 %T Performance Analytics by Means of the M5P Machine Learning Algorithm %U https://gitlab2.informatik.uni-wuerzburg.de/itc-conference/itc-conference-public/-/raw/master/itc31/fie19ITC31.pdf?inline=true %X Machine Learning (ML) has shown its capability to analyse, classify, and make predictions based on, large data sets, amongst others by means of decision trees. Network performance analysis and evaluation, on the other hand, focuses on finding and expressing qualitative, quantitative and preferably formal relationships between performance parameters. Due to the potential complexity of the latter, approximations that highlight main contributions and their orders of magnitude are of specific interest. Thereby, different parameter sub-spaces may imply different kinds of dependencies, sensitivities and approximation formulae, e.g. saturation, asymptotic and disruptive behaviours. Given such challenges, this work demonstrates the ability of the ML algorithm M5P to perform performance analytics by identifying approximations, together the applicable parameter sub-spaces, in a robust one-strike approach. where the detailed investigation of the obtained model trees provides valuable insights. We present, investigate and discuss a set of examples, spanning from impacts of parameters on user ratings, via modelling of user ratings over time, to post-analysis of analytically obtained performance results for approximations and asymptotic behaviour.

@inproceedings{fie19ITC31, abstract = {Machine Learning (ML) has shown its capability to analyse, classify, and make predictions based on, large data sets, amongst others by means of decision trees. Network performance analysis and evaluation, on the other hand, focuses on finding and expressing qualitative, quantitative and preferably formal relationships between performance parameters. Due to the potential complexity of the latter, approximations that highlight main contributions and their orders of magnitude are of specific interest. Thereby, different parameter sub-spaces may imply different kinds of dependencies, sensitivities and approximation formulae, e.g. saturation, asymptotic and disruptive behaviours. Given such challenges, this work demonstrates the ability of the ML algorithm M5P to perform performance analytics by identifying approximations, together the applicable parameter sub-spaces, in a robust one-strike approach. where the detailed investigation of the obtained model trees provides valuable insights. We present, investigate and discuss a set of examples, spanning from impacts of parameters on user ratings, via modelling of user ratings over time, to post-analysis of analytically obtained performance results for approximations and asymptotic behaviour.}, added-at = {2020-04-29T15:29:04.000+0200}, address = {Budapest, Hungary}, author = {Fiedler, Markus}, biburl = {https://www.bibsonomy.org/bibtex/217b286a3d8eb93965863e1a94454d104/itc}, booktitle = {31th International Teletraffic Congress (ITC 31)}, interhash = {d59f9415d814da090e5d043ffba90b7c}, intrahash = {17b286a3d8eb93965863e1a94454d104}, keywords = {itc itc31}, timestamp = {2020-04-30T18:18:45.000+0200}, title = {Performance Analytics by Means of the M5P Machine Learning Algorithm}, url = {https://gitlab2.informatik.uni-wuerzburg.de/itc-conference/itc-conference-public/-/raw/master/itc31/fie19ITC31.pdf?inline=true}, year = 2019 }