基于改进Shapley权力指数的特征选择算法
发布时间：2018-08-11 16:44
【摘要】：针对特征选择算法对不同类型的数据集性能不稳定的问题,提出一种基于概率模型与改进Shapley权力指数的通用特征选择算法.首先,计算特征对类簇表征与类簇判别的重要性值;然后,计算特征对类簇的不确定度;最终,合并特征的重要性值与不确定度,提取合适的特征.因为概率模型对数据类型、数据缺陷具有较好的鲁棒性,所以对不同的数据集获得了稳定、高性能的特征选择效果.基于人工合成数据与benchmark数据集的实验结果表明,本算法对不同的数据集保持了稳定的特征选择效果,优于其他算法.
[Abstract]:A general feature selection algorithm based on probabilistic model and improved Shapley power index is proposed to solve the problem of unstable performance of feature selection algorithm for different types of data sets. Firstly, the importance of feature to cluster representation and cluster discrimination is calculated; then, the uncertainty of feature to cluster is calculated; finally, the importance value and uncertainty of feature are merged to extract the appropriate feature. Because the probabilistic model is robust to data types and data defects, it can obtain a stable and high performance feature selection effect for different data sets. Experimental results based on synthetic data and benchmark datasets show that the proposed algorithm has a stable feature selection effect on different datasets and is superior to other algorithms.
【作者单位】： 镇江市高等专科学校装备制造学院;
【分类号】：TP301.6
，
本文编号：2177612
[Abstract]:A general feature selection algorithm based on probabilistic model and improved Shapley power index is proposed to solve the problem of unstable performance of feature selection algorithm for different types of data sets. Firstly, the importance of feature to cluster representation and cluster discrimination is calculated; then, the uncertainty of feature to cluster is calculated; finally, the importance value and uncertainty of feature are merged to extract the appropriate feature. Because the probabilistic model is robust to data types and data defects, it can obtain a stable and high performance feature selection effect for different data sets. Experimental results based on synthetic data and benchmark datasets show that the proposed algorithm has a stable feature selection effect on different datasets and is superior to other algorithms.
【作者单位】： 镇江市高等专科学校装备制造学院;
【分类号】：TP301.6
，
本文编号：2177612
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2177612.html
