Developing metric MDS techniques for the visualization and interpretation of customer data

With the growth of customer information systems, there is an increasing amount of customer data available to companies. There is a great need to be able to analyze and interpret these data. An important data analysis tool in the interpretation of data is that of visualization. This dissertation concentrates on one particular visualization technique, that of distance based metric multidimensional scaling (MDS). The dissertation concentrates on the development of metric MDS as a serious tool for visualization and data analysis.;The dissertation consists of four papers. Each paper concentrates on some aspect of the theory or application of distance based metric scaling. The underlying thread across all of these papers is the development of metric MDS as a tool for the analysis of large-scale, real world data sets.;The first paper develops a metric psi, based upon the Rand index, for the comparison and evaluation of dimensionality reduction techniques. This metric is designed to test the preservation of neighborhood structure in derived lower dimensional configurations and to compare different solution configurations for differences in structure.;The second paper is concerned with the performance of different distance metrics for high dimensional data mining and data analysis. The paper shows both the theoretical and empirical relationships between the cosine, correlation, and Euclidean metrics. The paper proposes that some of the performance difference between the cosine and correlation metrics and the Minkowski-p metrics, previously thought to be because of distance compression, is due to the inbuilt normalization of the cosine and correlation metrics.;In paper three, a method called DEMScale is introduced for large scale MDS. DEMScale can be used to reduce MDS problems into manageable sub-problems, which are then scaled separately. The DEMScale method is general, and is independent of MDS technique and optimization method.;The fourth paper develops an algorithm for modeling competitive market structure from auction data. The algorithm outputs measures of similarity and dissimilarity between products using auction bidding data. The paper shows how this similarity/dissimilarity information can be used to produce visualizations of product competition using MDS.