Offline decision tree-based microscale evolutionary algorithm for multimodal multiobjective optimization

Fangqing Liu

doi:10.1016/j.eswa.2026.133970

Return

Offline decision tree-based microscale evolutionary algorithm for multimodal multiobjective optimization

2026-08-19

PRE

DOI：10.1016/j.eswa.2026.133970

Original

Save

Abstract

En 中文

• Proposes a novel ODT-MEA algorithm for multimodal multiobjective optimization. • Predicts Pareto front types via offline decision trees mining population patterns. • Outperforms state-of-the-art methods via adaptive fitness evaluation allocation. Abstract Multimodal multiobjective optimization problems (MMOPs) present unique optimization challenges, characterized by intricate Pareto fronts and mutually conflicting objectives. While existing evolutionary algorithms perform well in single-Pareto-front scenarios, they often fail to address MMOPs effectively, primarily due to the lack of robust methods for predicting Pareto front types and adaptively allocating computational resources-two critical factors for high-efficiency optimization. To address this gap, this paper proposes a microscale evolutionary algorithm (MEA) tailored specifically for MMOPs, namely the Offline Decision Tree-based MEA (ODT-MEA). This method leverages offline-trained decision trees to predict Pareto front types in MMOPs by mining intrinsic population distribution patterns, which enables adaptive fitness evaluation scheduling to minimize redundant computations. Consequently, the proposed algorithm significantly enhances optimization efficiency in complex high-dimensional decision-making scenarios. We hypothesize that modeling population distribution patterns can precisely predict Pareto front types, thereby facilitating dynamic resource allocation to high-potential solution subspaces and improving the convergence and diversity of the obtained optimal solution sets. Experimental results demonstrate that ODT-MEA accurately predicts Pareto front types and outperforms state-of-the-art peer algorithms across widely used MMOP benchmark suites, which validates its superior efficiency and effectiveness in solving multimodal multiobjective optimization tasks.

Keywords:

Multimodal multiobjective optimization

Offline decision tree

Evolutionary algorithm

Computational intelligence