Local Landscape Patterns for Fitness Landscape Analysis

Almost all problems targeted by evolutionary computation are black-box or heavily complex, and their fitness landscapes usually are unknown. Selection of the appropriate search algorithm and parameters is a crucial topic when the landscape of a given target problem could be unknown in advance. Although several landscape features have been proposed in this context, examining a variety of landscape features is useful for problem understanding. In this paper, we propose a novel feature vector for characterizing the fitness landscape using the local landscape patterns (LLP). The proposed feature vector is composed by the histogram of the fitness patterns of the local candidate solutions. We extract the proposed LLP feature vector from well-known continuous optimization benchmark functions and BBOB 2013 benchmark set to investigate the properties of the proposed landscape feature and discuss about its effectiveness.