一种基于MDS的高维数据降维与可视化方法
首发时间:2017-05-03
摘要:降维与可视化是分析高维数据的有效手段。传统数据降维技术计算效率低,准确性较差,无法帮助分析者更深入理解和认识数据。因此,研究并实现了一种新的降维及可视化方法--最小二乘映射 (Least Squares Projection, LSP)方法。使用K中心点聚类算法选取每个簇中最有代表性的数据样本作为控制点;使用传统的降维方法--多维尺度分析 (Multidimensional Scaling, MDS),映射控制点的坐标;根据数据样本的邻域,建立相应的线性系统,从而计算出所有样本的降维结果。最后,利用多组数据验证了本方法的有效性,探究了本方法相比传统降维方法的优势,讨论了控制点的选取对结果的影响,并给出了本方法在教育信息中的具体应用。相比于传统方法,LSP方法能够在准确可视化数据的同时提高了计算效率。
For information in English, please click here
A New Method of Projecting and Visualizing High-dimensional Data Based on MDS
Abstract:Taditional multidimensional projection methods have low computational efficiency and poor accuracy, and these cannot be more indepth understanding and awareness of the data. Therefore, this paper studies a new multidimensional projection and visualization method called Least Squares Projection( LSP). The K-Medoid clustering algorithm classifies data into k clusters, meanwhile, we recognize the most representative data sample in each cluster as a control point. Then, the traditional method of dimensionality reduction called Multidimensional Scaling (MDS) is used to map the control points based on the relationships between different control points. and the linear system is built. Then, the coordinates of all data points are calculated. In the last of the article, multiple data sets are used to demonstrate the effectiveness of this method, explore the advantages of this method compared to traditional methods of dimension reduction, discuss the effects of selected control points on the results, and give an application in education information. Compared to traditional methods, LSP method can visualize data accurately and improve computational efficiency at the same time.
Keywords: High-dimensional data Dimension reduction K-Medoids MDS
论文图表:
引用

No.4727797922121149****
同行评议
勘误表
一种基于MDS的高维数据降维与可视化方法
评论
全部评论