Multidimensional multivariate information visualization using dynamic iconography

Large multidimensional multivariate (mdmv) datasets are now common in many fields. The data in various digital archives exist in a multitude of formats and are too voluminous to inspect or explore manually using numeric or tabular formats. This thesis proposes a new informational visualization approach for exploration of mdmv datasets to enhance data mining and knowledge discovery. The proposed technique for qualitative exploratory data analysis is to use dynamic iconography and tree icons. Visual iconography relies on the use of icons to visually map the multidimensional feature vectors of the dataset to three-dimensional representation to enable observers to effectively obtain qualitative understanding and insight. Conventional iconography such as stick figure visualization, however, is limited to two dimensions for graphical display and static icons are of limited use for mdmv datasets. We propose two enhancements through a repertoire of animations and through the use of tree icons that are more flexible and scalable three-dimensional icon maps. Three different types of icons were explored including propeller type motions and stylized humanoid gait, along with branching structures for tree icons, each with their own unique features, properties and capabilities to accommodate the various types of mdmv data. Three different datasets were collected and analyzed including multi-spectral Landsat data for landcover classification, handwritten digit recognition for machine learning and an e-commerce financial survey data for web assurance. The results using the StickViz software developed in this thesis using dynamic iconography demonstrates the utility, potential and promise of a new approach for understanding mdmv datasets.