Title: Study of meta-learning attempts to select algorithm for hard combinatorial optimisation problems

Authors: Neha Sehta; Urjita Thakar

Addresses: Department of Computer Engineering, Shri G.S. Institute of Technology and Science, Indore, India ' Department of Computer Engineering, Shri G.S. Institute of Technology and Science, Indore, India

Abstract: Given a large number of techniques to solve a problem, recommending the most appropriate for a specific case is defined as algorithm selection problem. Machine learning community identified this as learning task and called it meta-learning. It aims to identify a mapping from instance characteristics to algorithm performance. In this paper, review of meta-learning attempts for algorithm selection on different optimisation problems, is presented. Each meta-data tuple contains structural and other instance features along with target algorithm performance metrics. Solution quality, success probability, approximation ratio, algorithm ranking, run-time requirement are key performance measures. Supervised and unsupervised learning techniques have been applied effectively. Algorithm selection is considered valuable when a vast number of methods exhibit variable performance on different instances. The black-box approach to algorithm selection can be dropped if it is known how instance attributes relate to algorithm performance. In contrast to straightforward selection methods, meta-learning models lead to acquire high-quality solutions.

Keywords: machine learning; meta-learning; algorithm selection problem; ASP; optimisation problems.

DOI: 10.1504/IJIEI.2025.150111

International Journal of Intelligent Engineering Informatics, 2025 Vol.13 No.4, pp.461 - 486

Received: 09 Apr 2024
Accepted: 25 Aug 2024
Published online: 01 Dec 2025 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article