基于偏最小二乘回归的鲁棒性特征选择与分类算法

发布时间：2018-04-19 23:34

本文选题：偏最小二乘回归 + k近邻　；参考：《计算机应用》2017年03期

【摘要】：提出一种基于偏最小二乘回归的鲁棒性特征选择与分类算法(RFSC-PLSR)用于解决特征选择中特征之间的冗余和多重共线性问题。首先,定义一个基于邻域估计的样本类一致性系数;然后,根据不同k近邻(k NN)操作筛选出局部类分布结构稳定的保守样本,用其建立偏最小二乘回归模型,进行鲁棒性特征选择;最后,在全局结构角度上,用类一致性系数和所有样本的优选特征子集建立偏最小二乘分类模型。从UCI数据库中选择了5个不同维度的数据集进行数值实验,实验结果表明,与支持向量机(SVM)、朴素贝叶斯(NB)、BP神经网络(BPNN)和Logistic回归(LR)四种典型的分类器相比,RFSC-PLSR在低维、中维、高维等不同情况下,分类准确率、鲁棒性和计算效率三种性能上均表现出较强的竞争力。
[Abstract]:A robust feature selection and classification algorithm based on partial least squares regression (PLS) is proposed to solve the redundancy and multiple collinearity problems between features in feature selection. Firstly, a class consistency coefficient based on neighborhood estimation is defined, then conservative samples with stable local class distribution structure are selected according to different k-nearest neighbor KNN operations, and the partial least square regression model is established. Finally, the partial least squares classification model is established by using the class consistency coefficient and the optimal feature subset of all samples in terms of global structure. Five data sets of different dimensions are selected from UCI database for numerical experiments. The experimental results show that compared with four typical classifiers, support vector machine (SVM), naive Bayesian BP neural network (BPNN) and Logistic regression (LRSR), RFSC-PLSR is of low dimension and middle dimension. The classification accuracy, robustness and computational efficiency are highly competitive under different conditions such as high dimension.
【作者单位】：郑州大学电气工程学院;
【基金】：国家自然科学基金资助项目(U1304602,61473266,61305080) 河南省高等学校重点科研项目(15A120016)~~
【分类号】：TP301.6

【相似文献】