.png){kind=icon}
Return
MPEA-FS: A decomposition-based multi-population evolutionary algorithm for high-dimensional feature selection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2024.123296.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The challenge of high -dimensional feature selection (FS) lies in the search technique, which needs to consider both minimizing the size of feature subset and maximizing the classification accuracy. Recently, multi -objective evolutionary algorithms (MOEAs) have shown excellent performance in solving FS tasks. However, most existing MOEAs struggle to effectively balance these two conflicting objectives when solving high -dimensional FS tasks. To cover this issue, in this paper, a decomposition -based multi -population evolutionary algorithm is proposed, called MPEA-FS. In the initialization stage, a multi -population generation strategy (MGS) based on feature weights (MGS) was adopted, with each population corresponding to a search subspace, which can improve the algorithm's search ability. During the evolutionary stage, an external population is designed to integrate the excellent feature subsets of multiple populations to achieve knowledge sharing between them. In addition, to reduce the feature dimension, a feature reduction strategy (FRS) is proposed, which can remove unimportant features while maintaining classification accuracy. Extensive experiments are performed on 13 high -dimensional datasets, and the proposed algorithm is compared with 5 state-of-the-art FS methods proposed in the last 3 years on multiple metrics. The experimental results indicate that MPEA-FS can achieve higher classification accuracy on most datasets, and the number of selected features is also competitive.
Keywords:
Classification
High-dimensional feature selection
Decomposition-based method
Multi-population evolutionary
Knowledge sharing
Journal
IF:
7.5
Papers:
3.0W
Citations:
10.2W
Organization
Cited Papers
ShareSaveMPF-FS: A multi-population framework based on multi-objective optimization algorithms for feature selection.png)
APPLIED INTELLIGENCE
APPLIED INTELLIGENCEIF3.5
PREAI
ShareSaveShareSave
Multiobjective evolutionary algorithms: A comparative case study and the Strength Pareto approach
PREAI
ShareSaveShareSaveShareSave
View more