Improving the Performance of Genetic Algorithms for Combinatorial Optimization Using Machine Learning for Knowledge Transfer

George Mweshi, Nelishia Pillay

2024

Abstract

This study investigates improving the performance of genetic algorithms applied to the solution space using machine learning and knowledge transfer. Genetic algorithms are powerful techniques that have been successfully used to explore various problem spaces, such as solution space, program space, and heuristic space. Recently, researchers have found that transferring knowledge between these spaces can significantly enhance the quality of solutions and reduce computational costs. While this transfer of knowledge works well in program and heuristic spaces due to their indirect nature, it is more challenging in the solution space. This is because each problem in the solution space has its own unique representation, making it difficult to transfer knowledge effectively. This study explores how machine learning, specifically using classifiers, can help bridge this gap and facilitate knowledge transfer between different solution spaces. We train two classifiers, namely, Support Vector Machines and Random Forests, using data consisting of fitness landscape measures from a source genetic algorithm to determine if a chromosome is a local optimum or not. This information is then used during the execution of a target genetic algorithm to identify and remove potential local optima from the population. We tested this approach on two challenging optimization problems: the examination timetabling problem (ETP) and the capacitated vehicle routing problem (CVRP). Our results show that this method provides statistically significant improvements over genetic algorithms that do not use knowledge transfer, both in terms of solution quality and computational efficiency. Moreover, we found that random forests were more effective than support vector machines for transferring knowledge between the source and target genetic algorithms.

Paper Citation

in Harvard Style

Mweshi G. and Pillay N. (2024). Improving the Performance of Genetic Algorithms for Combinatorial Optimization Using Machine Learning for Knowledge Transfer. In Proceedings of the 16th International Joint Conference on Computational Intelligence - Volume 1: ECTA; ISBN 978-989-758-721-4, SciTePress, pages 363-374. DOI: 10.5220/0013084200003837

in Bibtex Style

@conference{ecta24,
author={George Mweshi and Nelishia Pillay},
title={Improving the Performance of Genetic Algorithms for Combinatorial Optimization Using Machine Learning for Knowledge Transfer},
booktitle={Proceedings of the 16th International Joint Conference on Computational Intelligence - Volume 1: ECTA},
year={2024},
pages={363-374},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013084200003837},
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 - Improving the Performance of Genetic Algorithms for Combinatorial Optimization Using Machine Learning for Knowledge Transfer
SN - 978-989-758-721-4
AU - Mweshi G.
AU - Pillay N.
PY - 2024
SP - 363
EP - 374
DO - 10.5220/0013084200003837
PB - SciTePress