基于特征相关的偏最小二乘特征选择方法-ChinaXiv.PDF

/article/02-2019-04-006.html 计算机应用研究 基于特征相关的偏最小二乘特征选择方法 a a a a b a 曾青霞 ，杜建强 ，朱志鹏 ，聂 斌 ，余日跃 ，喻 芳 (江西中医药大学 a. 计算机学院; b. 药学院, 南昌 330004 ) 摘 要：针对传统的偏最小二乘法只考虑单特征的重要性以及特征之间存在冗余和多重共线性等问题，将特征之间的 统计相关性引入到传统的偏最小二乘分析中，构造了一种基于特征相关的偏最小二乘模型。首先利用特征相关度对特 征进行评估预选出特征组，然后将其放入偏最小二乘模型中进行训练，评估该特征组是否可取。结合前向贪心搜索策 略依次评价候选特征，并选中使目标函数最小的候选特征加入到已选特征。分别采用麻杏石甘汤君药止咳、平喘和UCI 数据集进行分析处理，实验结果表明，该特征选择方法能较好寻找较优的特征组。 关键词：中医药信息；偏最小二乘法；特征相关；特征选择 中图分类号：TP PLS feature selection method based on feature correlation 1 v a a a b A A 3 Zeng Qingxia , Du Jianqiang , Nie Bin , Yu Riyue , Yu Fang , Huang Canyi 6 0 (School of Computy Jiangxi University of Traditional Chinese Medicine , Nanchang 330004, China) 0 0 Abstract: The traditional partial least squares method only considers the importance of single features and it exists the . 5 redundancy and multicollinearity among the features . The statistical correlation between features is involved into the traditional 0 partial least squares analysis, and constructed the model of PLS feature selection based on feature correlation. Firstly, the feature 8 1 group is pre-selected by using of the feature relevance, and then put into the partial least squares model for training to assess 0 whether the feature group is desirable . Combining with the greedy search strategy, the candidate features are evaluated one by 2 : one, and the candidate features with the smallest objective function are added to the selected features . Respectively, usi