On the “fairness” of search-based software engineering: Investigating the capability of a scalarizing-based function to control the software metrics’ influence on optimization process

Search-Based Software Engineering (SBSE) aims to transform Software Engineering (SE) problems into search problems by defining a fitness function that guides the search for an optimal or sub-optimal solution. However, designing a fitness function that provides equitable relevance (or weight) to every SE metric associated with an SBSE problem is an assumption that may be challenging. This issue derives from the several properties related to SE metric value domains that can induce the search process to privilege specific metrics over others, misleading to suboptimal outcomes. To deal with this problem, this work proposes a mathematical model based on the scalarization function concept to better control each metric’s relevance in the search process. Our empirical study comprises two computational experiments. The first experiment aimed to evaluate the proposed scalarizing-based approach’s control capability over the SE metrics in a scenario where all metrics should have the same relevance, while the second experiment covers the scenario where metrics do not necessarily should have the same relevance. The results demonstrate the importance of properly considering the impact, nature, and value range of SE metrics in the search process and the effectiveness of the proposed model in controlling SE metric relevance in different scenarios. This research makes three significant contributions. Firstly, we empirically highlight the importance of properly considering the relevance of individual SE metrics in the search process. Secondly, we propose a generic mathematical model based on scalarizing functions to cope with the normalization process and can be applied to a wide range of SBSE problems. Finally, we show that the our scalarizing approach is capable of guiding search-based process not only in the scenario where all metrics relevance must be equal, but also in the variation of the relevance alongside the optimization process, which is quite important for the design of fitness functions in SBSE.