.png){kind=icon}
返回
Reinforced feature selection using Q-learning based on collaborative agents
.png){kind=icon}
PRE
AI
DOI：10.1007/s13042-023-01869-8.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Reinforced feature selection (RFS) applies reinforcement learning to feature selection, which can continue to learn the procedure of feature selection and then effectively find the optimal feature subset from original features. Q-learning with Fisher score (QLFS), a kind of filtering RFS algorithm, adopts the Fisher score to construct the internal reward and then achieves good performance on large-scale data sets. However, it is hard for QLFS to automatically determine how many features should be selected. To remedy it, this paper proposes a novel RFS method, called Q-learning based on collaborative agents (QLCA). To implement automatic feature selection, QLCA adopts two agents to collaboratively learn the optimal strategies. One agent, called selection agent, is used to select features, where the Fisher scores of feature subsets are taken as the reward; while the other agent, called classification agent, is to determine how many features should be chosen, where the classification performance of a classifier on selected features is regarded as the reward. The selection agent provides feature subsets for the classification agent, while the classification agent gives the classification performance to the selection agent. In doing so, these two agents work well together in QLCA. Extensive experiments were conducted on several small and large real-world data sets. According to statistical analysis on experimental results, QLCA is significantly better than QLFS. Findings indicate that QLCA can effectively and automatically perform feature selection.
Keyword:
Reinforced feature selection
Q-learning
Fisher score
Agent
期刊
IF:
2.7
论文数:
3.2K
被引数:
5.6K
机构
引用论文
.png){kind=icon}
Estrogen Receptor-β Action in the Ovary, but Not in the Hypothalamic-Pituitary Axis, Is Required for the Generation of Normal Luteinizing Hormone Surges in Mice.雌激素受体-β在卵巢中的作用，但在下丘脑-垂体轴中并非如此，对于小鼠正常黄体生成素峰值的产生是必需的。
PREAI
分享收藏The Use of Ranks to Avoid the Assumption of Normality Implicit in the Analysis of Variance使用秩来避免方差分析中隐含的正态性假设
PREAI
分享收藏
查看更多内容