Fitness Landscape Analysis of Circles in a Square Packing Problems

Fitness landscape analysis provides insight into the structural features of optimization problems. Landscape analysis techniques have been individually shown to capture specific continuous landscape features. However, results are typically for benchmark and artificial problems, and so the ability of techniques to capture real-world problem structures remains largely unknown. In this paper we experimentally examine and compare the ability of length scale analysis, dispersion, fitness distance correlation, information content, partial information content and information stability to characterise and distinguish instances of circle packing problems. Circle packing problems are an important abstraction of many real-world problems such as container loading, facility dispersion and sensor network layout problems. Experiments on incrementally scaled packings show that while all of the techniques provide some problem insight, only length scale analysis and information stability were able to clearly differentiate problem instances.