基于注意力机制和HSIC Lasso的特征选择算法  

Feature Selection Algorithm based on Attention Mechanism and HSIC Lasso

作　　者：胡鹏程 余聪 刘启枫 石太贵 刘汉明[1] HU Pengcheng;YU Cong;LIU Qifeng;SHI Taigui;LIU Hanming(School of Mathematics and Computer Science,Gannan Normal University,Ganzhou 341000,China)

机构地区：[1]赣南师范大学数学与计算机科学学院,江西赣州341000

出　　处：《赣南师范大学学报》2024年第6期33-38,共6页Journal of Gannan Normal University

基　　金：江西省教育厅科学技术研究项目(GJJ2201203)。

摘　　要：特征选择作为高维数据预处理的重要方法,广泛应用于机器学习与数据挖掘任务并取得了良好的效果.然而,随着高维、海量的大数据时代到来,现有的特征选择算法面临着特征选择后所取得的分类准确率不甚理想等方面的挑战.本文提出了一种全新的特征选择算法AHSIC Lasso,通过运用注意力机制寻找特征之间的关联性,根据SoftMax计算特征权重并对数据进行加权,然后基于加权结果,使用HSIC Lasso方法对加权后的数据作特征选择.实验结果表明,本文提出的算法在精度上相比于传统的特征选择算法有较大的提高.Feature selection,as an important method for preprocessing high-dimensional data,is widely used in machine learning and data mining tasks as well as approved by its well performance.However,in today's era of big data,the existing feature selection algorithms are facing challenges such as the classification accuracy that the selected features can provide.This study proposes a new feature selection algorithm,AHSIC Lasso.It uses the attention mechanism to find the correlation between features,calculates feature weights according to SoftMax and weights the data.And then,itemploys HSIC Lasso method to select features from the weighted data.Experimental results show that the algorithm has a higher accuracy than the traditional feature selection algorithms.

关 键 词：特征选择 数据降维 注意力机制 HSIC Lasso AHSIC Lasso 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]