一种基于图的线性判别分析方法

线性判别分析(LDA)作为全局性降维的方法,在处理局部性边缘点的问题上存在不足,可能会导致边缘点的误分。针对该问题,提出一种新的降维方法,该方法基于图学习的思想,重新构造图,使得同类之间向中心靠拢的同时,不同类的K个近邻点远离该类中心。这样,高维数据在嵌入低维的过程中保持了样本的局部边缘点的特性,从而保证了边缘点的正确分类。通过在UCI数据集和人脸数据库中实验,结果表明本方法的有效性。

Linear discriminant analysis method based on graph

Most dimension reduction methods can be attributed to the context of map learning method, Linear Discriminant Analysis (LDA) also belongs to one of them. Generally speaking,we believed that the LDA has a good effect in dealing with classification problems, However, as global dimensionality reduction method, LDA has deficiencies in dealing with localized edge point deficiencies, and it may lead to misclassification of edge points. To solve this problem this paper proposes a new dimensionality reduction method, the idea of this method is based on the map learning method. we reconstruct the map, make closer to the center of its kind between the same class, at the same times, we make K nearest neighbors to stay away from such centers in different classes. In this way, the high-dimensional data has kept the characteristics of the local edge points of the sample in the process of embedded low-dimensional, and it ensures the correct classification of edge points. We do experiments on UCI data sets and face database, the results show that the effectiveness of the method.