.png){kind=icon}
.png){kind=icon}
Return
Reinforcement learning guided auto-select optimization algorithm for feature selection
.png){kind=icon}
PRE
AI
DOI：10.1016/j.eswa.2024.126320.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Feature selection (FS) is increasingly important in classification tasks. Although metaheuristic algorithms have been extensively utilized for FS problems, they share the common drawbacks that the quality of the initial population is challenging to guarantee and easily trapping into stagnation. To handle these drawbacks, a reinforcement learning guided auto-select optimization algorithm (RLAS) is proposed for FS problems. To be more specific, a median initialization strategy is designed to synthetically utilize the advantages of several filter FS methods, which plays a virtual role in ensuring the initial population quality. In particular, a Q-learning guided auto-select mechanism is developed to choose the optimal module based on the current search status. Finally, a feature importance and random-based grouping mechanism is constructed to promote the search ability by detecting interactions among features. We evaluate the performance of the RLAS on 20 benchmark datasets through several well-known metrics. The experiment outcomes reveal that the RLAS is significantly better than the typical and advanced algorithms, thereby proving its efficiency while solving FS problems.
Keywords:
Feature selection
Data mining
Reinforcement learning
Metaheuristic algorithms
Journal
IF:
7.5
Papers:
2.9W
Citations:
10.2W