Sample-Cluster-Select: A new framework to obtain diverse approximate solutions of combinatorial optimization problems

Abstract

When solving real-world problems, practitioners often hesitate to implement solutions obtained from mathematical models, especially for important decisions. This hesitation stems from practitioners' lack of trust in optimization models and computational results. To address these challenges, we propose Sample-Cluster-Select (SCS) for solving practical combinatorial optimization problems under the assumption of potentially acceptable solution set. SCS first samples the potential solutions, performs clustering on these solutions, and selects a representative solution for each cluster. SCS aims to build trust by helping users understand the solution space through multiple representative solutions, while simultaneously identifying promising alternatives around these solutions. We conducted experiments on randomly generated instances, comparing SCS against multi-start local search and $k$-best algorithms where efficient methods exist, or evolutionary algorithms otherwise. The results show that SCS outperforms multi-start local search and $k$-best algorithms in most cases, while showing competitive performance against evolutionary algorithms, though not surpassing some of their variants. Most importantly, we found that the clustering approach provides insights into solutions that are difficult to obtain with existing methods, such as local structures of similar potential solutions and neighboring solutions of representative solutions. Thus, our approach helps practitioners understand the solution space better, thereby increasing their confidence in implementing the proposed solutions.