Evaluating Genetic Algorithms through the Approximability Hierarchy

Alba Muñoz,Fernando Rubio

DOI: https://doi.org/10.1016/J.JOCS.2021.101388

2024-02-01

Abstract:Optimization problems frequently appear in any scientific domain. Most of the times, the corresponding decision problem turns out to be NP-hard, and in these cases genetic algorithms are often used to obtain approximated solutions. However, the difficulty to approximate different NP-hard problems can vary a lot. In this paper, we analyze the usefulness of using genetic algorithms depending on the approximation class the problem belongs to. In particular, we use the standard approximability hierarchy, showing that genetic algorithms are especially useful for the most pessimistic classes of the hierarchy

Neural and Evolutionary Computing

What problem does this paper attempt to address?

This paper discusses the effectiveness of Genetic Algorithms in solving NP-hard problems, depending on the approximability class of the problem. The author points out that although finding the optimal solution for NP-hard problems is equally difficult, finding good approximate solutions varies in difficulty. The paper analyzes the performance of Genetic Algorithms in different approximability classes through a standard approximation hierarchy. The experimental results show that for problems with poorer approximation capabilities, Genetic Algorithms perform better. Specifically, when the problem belongs to the best approximability class (such as PTAS and FPTAS), Genetic Algorithms do not have a clear advantage over specific problem heuristic methods; while in the worst approximability class (such as Exp-APX), Genetic Algorithms can achieve better results than heuristic methods.

The paper first introduces the concept of NP-hard problems, then defines the performance ratio of approximation algorithms, and divides NP-hard optimization problems into different approximability classes, such as Exp-APX, Poly-APX, Log-APX, APX, PTAS, and FPTAS. Then, the author conducted experimental studies on representative problems from each approximability class, comparing the performance of Genetic Algorithms with specific problem heuristic methods. The experiments showed that the worse the approximation capability of the problem, the more significant the improvement in the performance of Genetic Algorithms.

In addition, the paper proposes a basic implementation of Genetic Algorithms and compares them using standard benchmarks for each problem. In all experiments, the performance of Genetic Algorithms improves as the difficulty of the approximation problem increases. Although there have been many studies on the performance of Genetic Algorithms, there is no work reported on systematically evaluating their effectiveness based on the approximation hierarchy.

In conclusion, this paper reveals the applicability of Genetic Algorithms in dealing with NP-hard problems in different approximability classes through empirical studies, providing guidance for selecting appropriate optimization methods.