A0330
Title: Clustering based multidimensional scaling for mixed data
Authors: Mika Sato-Ilic - University of Tsukuba (Japan) [presenting]
Abstract: Multidimensional scaling (MDS) is a well-known method for dimensional reduction and visualization of multidimensional data. Clustering-based MDS is proposed for data of the mixed types, which comprises numerical and categorical data regarding quantitative and qualitative variables. For this type of data, the main difficulty for the treatment of data depends on the difference in quantity and quality of intrinsic information contained in the data. The amount of data information is larger for the numerical data; therefore, traditionally, the transformation from the numerical data to categorical data by aggregating the data information for the categorical data has been used. However, in this case, complex methods are needed for the aggregation of the transformation. Therefore, a simple technique which can treat both data types through the same data projection into the lower dimensional space is proposed. This method includes the classification structure obtained by using fuzzy clustering; the categorical data part of the mixed data can be summarized and expresses the simple relationship with the objects of the numerically obtained data part of the mixed data.
