An Introduction to Fitness Landscape Analysis and Cost Models for Local Search

Abstract

Despite their empirical effectiveness, our theoretical understanding of metaheuristic algorithms based on local search (and all other paradigms) is very lim- ited, leading to significant problems for both researchers and practitioners. Specifi- cally, the lack of a theory of local search impedes the development ofmore effective metaheuristic algorithms, prevents practitioners from identifying the metaheuris- tic most appropriate for a given problem, and permits widespread conjecture and misinformation regarding the benefits and/or drawbacks of particular metaheuris- tics. Local search metaheuristic performance is closely linked to the structure of the fitness landscape, i.e., the nature of the underlying search space. Consequently, un- derstanding such structure is a first step toward understanding local search behavior, which can ultimately lead to a more general theory of local search. In this paper, we introduce and survey the literature on fitness landscape analysis for local search, placing the research in the context of a broader, critical classification scheme de- lineating methodologies by their potential to account for local search metaheuristic performance.