Evaluation of Genotypic Diversity Measurements Exploited in Real-Coded Representation
This work addresses a methodological gap for researchers in evolutionary computation, but it is incremental as it builds on prior calls for validation without introducing new measures.
The study tackled the problem of validating genotypic diversity measures (GDMs) used in evolutionary algorithms by proposing a framework with three requirements, and found that none of the four evaluated GDMs complied with all requirements, highlighting the difficulty of proper diversity evaluation.
Numerous genotypic diversity measures (GDMs) are available in the literature to assess the convergence status of an evolutionary algorithm (EA) or describe its search behavior. In a recent study, the authors of this paper drew attention to the need for a GDM validation framework. In response, this study proposes three requirements (monotonicity in individual varieties, twinning, and monotonicity in distance) that can clearly portray any GDMs. These diversity requirements are analysed by means of controlled population arrangements. In this paper four GDMs are evaluated with the proposed validation framework. The results confirm that properly evaluating population diversity is a rather difficult task, as none of the analysed GDMs complies with all the diversity requirements.