Multivariate Data Visualization: Correspondence Analysis, Classical and Robust Singular Value Decomposition and Depth based Approach

Now a day’s Data Mining is one of the challenging area in statistics as well as in computer science. Data visualization is one of the most important parts in Data Mining. We can only visualize two or three dimensional data but for Data Mining much more than three dimensions is usual rather than exception. So data reduction is very important for visualization of multivariate data. For this reason, in this article I would like to introduce four well known, effective and sophisticated scientific data reduction methods, correspondence analysis, singular value decomposition (SVD), robust singular value decomposition (RSVD) and depth for data visualization, pattern recognition and outlier detection. Since these techniques are used for data reduction technique so by using these techniques we can visualize data taking only two or three dimensions that maximize the total variation of data. In many cases two or three singular values or eigen values cannot explain most (greater than 80%) of the variation of data. In that case we can use L1 depth, half space depth and kernel based depth for visualizing data. . In this paper we have used four well known real dataset (fisher’s iris data, Wisconsin breast cancer data, Glass identification data and Seeds data) for visualization and also rigorously explained the results on the motion of the aforementioned techniques.