A software measurement framework guided by support vector machines

The quality of software engineering has always been of high importance for many actors. With the complexity of the platforms and its components, this is nowadays becoming crucial at each level in order to detect the eventual defects. Due to that complexity, the current measurement and analysis processes become heavier. Indeed, either for runtime monitoring, QoE, mobile gaming or simply for systems development, the software measurements tasks have to be fine-grained, 'greenable' and distributed. This work aims at improving the software monitoring processes and its analysis. Based on a learning-aided analysis, we intend to suggest and select metrics that should be applied at runtime to increase the quality of the measurement plan and to target metrics that could raise relevant information on the measureand. Our approach proposes a data model that allows highlighting the monitored activity of a characteristic according to the data values of the model. We focus on complex metrics that are formally modeled using the OMG standard SMM. Some experiments are performed to exemplify our methodology.