Improvement of design anti-pattern detection with spatio-temporal rules in the software development process

In [1] we presented a framework for mining spatio-temporal rules in the software development process. The rules are based on specific relations between structures of the source code which relate both to spatial (e.g. a direct call between methods of two classes) and temporal dependencies (e.g. one class introduced into the source code before the other) observed in the process. To some extent, spatio-temporal rules allow us to predict where and when certain design anti-patterns will appear in the source code of a software system. This paper presents how, with slight modifications, such framework can be used to improve the quality of detecting a few popular design anti-patterns, such as Blob, Swiss Army Knife, YoYo or Brain Class. In the proposed method, we not only check the structure of a piece of the source code, but we also analyse its spatio-temporal relations. Only on the basis of the two analyses can we decide if the given piece of code is an anti-pattern. Experimental validation shows that the addition of spatio-temporal perspective improves detection of anti-patterns by 4% in terms of F-measure.