Information Visualization of Multivariate Data
Contact: Jimmy Johansson
Visualization of multivariate data is one of the major challenges in information visualization. Such data is frequently becoming large and hard to manage but also, more importantly, is very difficult to represent in an understandable way. Existing representations have limitations in that they do not scale well with increasing data sizes, resulting in visual clutter or interactivity loss.
This project will research fundamental techniques for interactive visual representation of large, complex, multivariate data sets with a focus on visual identification and presentation of inter-related patterns. Application areas where such techniques have great potential are, for example, visualization of diagnosable properties, climate change and process control.
Regardless of application area or visualization technique used, the issue of visual clutter, due to too much data, needs to be considered. Even if the overall goal is to extract patterns, an initial visual representation of the overall structure of the data is often necessary in order to choose a starting point for the analysis. To avoid visual clutter a data mining algorithm is often applied to reduce the size of the data. During this step it is important that the overall structure of the data is retained so that the analysis result is still valid. Within the scope of this project new quality metrics will be developed that considers different aspects of structure in data.
A key challenge in information visualization research is the consideration of user prerequisites for dealing adequately with, and making sense of the visual representations offered. This project will therefore include significant evaluation of the developed techniques through both qualitative and quantitative studies.