A Comparison Between Hierarchical and Non-Hierarchical Software Clustering

Heuristic search algorithms have been applied in various areas of Software Engineering, such as requirements prioritization, software architecture improvement, and time and cost planning. The research field of Search-based Software Engineering (SBSE) proposes the use of heuristic search techniques to solve Software Engineering challenges, which are described as optimization problems. One of the challenges frequently analyzed from the SBSE perspective is the Software Module Clustering (SMC) problem. This problem involves distributing a software project’s component modules (or classes) among larger structures called clusters (or packages). Practice and experimental studies have shown a strong correlation between poor distributions and the presence of component failures. The main factors used to assess the quality of distributions are coupling and cohesion. However, research shows that optimization based on metrics that capture structural characteristics does not result in an adequate distribution of classes in packages from the point of view of software developers. One of the reasons that might have led to such limited results is that the algorithms produce non-hierarchical distributions of classes into packages. In this study, we executed the same greedy algorithm using two different fitness functions: one that does not consider hierarchy and the other that does. We wanted to measure whether they produced different results regarding authoritativeness. We found indications that a hierarchy-based approach produces solutions closer to those software developers propose.