An incremental random walk algorithm for sampling continuous fitness landscapes

Abstract

Fitness landscape analysis (FLA) is used to mathematically characterize optimization problems. It is critical to get appropriate sampling points using random walk (RW) algorithms to perform FLA on continuous optimization problems. However, most of the existing approaches face the problems of poor robustness of the RW algorithm and inadequate coverage of the fitness landscape. In this study, an incremental random walk (IRW) algorithm is proposed to sample the fitness landscape for continuous optimization problems. IRW includes two major improvements: (i) An incremental perturbation mechanism is proposed to generate perturbation variables to enhance the diverse distribution of sampling points. (ii) The problem of getting trapped in the local region is alleviated by using the mirrored boundary handling method. The experimental results tested on various cases demonstrate the excellence of IRW in terms of coverage. Moreover, IRW has shown superior reliability at different problem dimensions on other different benchmark functions. Consequently, IRW is well-suited as an alternative for sampling continuous fitness landscapes under a variety of problem dimensions.