Quantile and mean value measures of search process complexity

Performance measures of metaheuristic algorithms assess the quality of a search process by statistically analysing its performance. Such criteria serve two purposes: they provide the verdict on which algorithm is better for what task, and they help applying an algorithm on a given task in the most effective way. The latter goal may be achieved by an appropriate restart strategy of the search process. Furthermore, these criteria are traditionally based on analysis of the search step mean value. Our aim is to elaborate the mean value analysis as well, but via a novel and more general quantile-based analytic approach, which can be used to define new measures. We prove and demonstrate this purpose on three quantile-based performance measures.