Sunday 02 February 2025
The quest for optimisation has been a long-standing challenge in various fields, from engineering and economics to biology and medicine. In recent years, researchers have been tackling this problem by developing new algorithms that can efficiently solve complex multi-objective problems. These problems involve finding the best solution among multiple competing objectives, such as maximising profit while minimising cost.
A team of scientists has made significant progress in this area by testing various algorithms on real-world datasets from feature selection and location selection. Feature selection is a crucial step in machine learning, where the goal is to identify the most relevant features from a large dataset that can accurately predict a target outcome. Location selection, on the other hand, involves finding the best spot for a facility or infrastructure, taking into account multiple criteria such as distance, cost, and accessibility.
The researchers compared eight state-of-the-art algorithms with three classic methods on eight feature selection datasets and four location selection datasets from Guangzhou, China. The results showed that each algorithm performed well on certain problems but struggled on others. For instance, one algorithm excelled at finding equivalent feature subsets in feature selection problems, while another algorithm outperformed the rest in location selection problems.
The team also found that algorithms designed for multimodal multi-objective optimisation (MMOP) – a type of problem where multiple objectives are conflicting and there is no single optimal solution – performed better than those not specifically designed for MMOP. This suggests that these algorithms may be more effective at solving real-world problems, which often involve complex trade-offs between competing objectives.
The researchers identified several key factors that influence the performance of MMOP algorithms, including their ability to preserve local Pareto fronts (solutions that are optimal in a specific region) and maintain diversity in the decision space. They also found that algorithms using ring topology – where each individual communicates with its neighbours in a circular pattern – performed better than those without this structure.
The study highlights the importance of understanding how different MMOP algorithms work on real-world problems, rather than relying solely on benchmark functions or simulations. By developing more effective algorithms and adapting them to specific problem domains, researchers can unlock new possibilities for solving complex optimisation challenges.
Cite this article: “Efficient Solving of Multi-Objective Problems through Advanced Algorithms”, The Science Archive, 2025.
Multi-Objective Optimization, Algorithms, Feature Selection, Location Selection, Machine Learning, Multimodal Multi-Objective Optimisation, Mmop, Pareto Fronts, Decision Space, Ring Topology
