Fitness Histograms of Expert-Defined Problem Classes in Fitness Landscape Classification

Vojtěch Uher, Pavel Krömer

2024

Abstract

Various metaheuristic algorithms can be employed to find optimal or sub-optimal solutions for different problems. A fitness landscape (FL) is an abstraction representing a specific optimization task. Exploratory landscape analysis (ELA) approximates the FL by estimating its features from a limited number of random solution samples. Such ELA features help in estimating the properties of the FL and ultimately aid the selection of suitable optimization algorithms for problems with certain FL characteristics. This paper proposes using a normalized histogram of fitness values as a simple statistical feature vector for representing FLs. These histograms are classified using various classifiers to evaluate their effectiveness in representing different problems. The study focuses on 24 single-objective benchmark problems, grouped into five expert-defined classes. The performance of several classifiers is compared across different problem dimensions and sample sizes, emphasizing the impact of different sampling strategies and the number of histogram bins. The findings highlight the robustness of histogram representation and reveal promising experimental setups and relationships.

Paper Citation

in Harvard Style

Uher V. and Krömer P. (2024). Fitness Histograms of Expert-Defined Problem Classes in Fitness Landscape Classification. In Proceedings of the 16th International Joint Conference on Computational Intelligence - Volume 1: ECTA; ISBN 978-989-758-721-4, SciTePress, pages 205-213. DOI: 10.5220/0012923900003837

in Bibtex Style

@conference{ecta24,
author={Vojtěch Uher and Pavel Krömer},
title={Fitness Histograms of Expert-Defined Problem Classes in Fitness Landscape Classification},
booktitle={Proceedings of the 16th International Joint Conference on Computational Intelligence - Volume 1: ECTA},
year={2024},
pages={205-213},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012923900003837},
isbn={978-989-758-721-4},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 16th International Joint Conference on Computational Intelligence - Volume 1: ECTA
TI - Fitness Histograms of Expert-Defined Problem Classes in Fitness Landscape Classification
SN - 978-989-758-721-4
AU - Uher V.
AU - Krömer P.
PY - 2024
SP - 205
EP - 213
DO - 10.5220/0012923900003837
PB - SciTePress