Is it a Real Code Smell to be Removed or not?

Several kinds of smells have been defined in the literature, but no formal definition exists. Hence, the code smells detection tools provide different results, because the smells can be interpreted and detected in different ways. Moreover, other aspects related to the domain of the system under analysis, or its size and other design features are not usually taken into account. For these reasons the detected smells may not always be real smells and symptoms of problems to be solved. In this paper we address this question on the evaluation of smells. We focus our attention on the detection of two smells, God Class and Data Class on different systems, with the aim to identify possible filters to be used to improve or refine the detection rules of these smells. By identifying only real smells, the refactoring effort to inspect and remove them is reduced.