Detection of feature freezes using clustering algorithms

Software development is complex and not easy to quantify. An approach to quantify certain aspects of a software and software development are metrics. Software metrics map a possibly abstract attribute of a software to a value. To analyze the software project itself, is is possible to use metric data, that is measured at different versions of a project. The aim of this thesis is to determine whether such metric data can be used to analyze software projects using machine learning techniques. To obtain this kind of metric data, a way to mine metric data about different versions of a software from its archive is shown as part of this thesis. The aim is to detect a feature freeze using only this metric data. To do this, the k-means clustering algorithm is used, to divide the measured versions into those, that took place before the feature freeze and those that were afterwards. The experiment successfully detected a feature freeze in the tests candidates.