Feature space for statistical classification of Java source code patterns

To develop a reliable statistical classifiers of Java source code patterns, a feature space has to be developed and thoroughly examined as there are little general recommendations, such as in the field of image processing. This paper deals with development and evaluation of such feature space. Current version of feature space consisting of four categories and forty features is presented. Moreover, since feature collection from a source code is a non-trivial task, method of data acquisition with help of newly constructed domain specific language is given. Another issue that has to be solved is determination of structure of particular patterns, as their implementation can vary with different software projects. Straightforward patterns may have little explanatory power for project´s architecture, however it could be demanding to detect more abstract ones. In addition, the proposed patterns should meet standards recognized by the software engineering community.