Ronak Etemadpour

• This thesis presents experimental results associated with the human factors aspects of visualizing multidimensional data. Visual exploration of multidimensional data typically requires projection onto lower-dimensional representations. A large number of possible projections has been proposed in the recent years. The analysis of multidimensional data has been studied in various research areas for many years and different quality measures have been introduced to help in the interpretation of sets of points in a multidimensional space. Different measures deliver different visual metrics to evaluate the best views of multidimensional datasets that do not necessarily match the expectations of human perception. This thesis contributes toward this goal by providing some perceptual guidelines through investigating different projections. Learning more about the effectiveness of projection layouts from a user’s perspective is an important step towards consolidating their role in supporting visual analytics tasks. Such tasks often involve detecting and correlating clusters. This thesis supports discovering certain characteristics that change the visual attention and cognitive process for finding clusters, patterns, outliers, and relationships. The main approach consists of three major steps according to three fundamental parts of the visualization pipeline: Data preparation, Mapping, and Rendering. The visualization pipeline describes the (step-wise) process of creating visual representations of data. A good visualization guides the observers’ attention to the relevant aspects of the representation. To achieve this aim, correlations and connections between human perception and different visualization steps are investigated. At the first stage, the supporting visual analytics tasks are found through understanding the data set and learning more about domain-specific issues. This helps to improve decision process when making decisions becomes hard for selecting from alternative multidimensional data representations. In the second stage of our study, projection methods representative of different approaches are considered that generate different layouts of multidimensional data in a lower-dimensional visual space. Mapping or geometry extraction is the main core of the visualization process. In the case of high-dimensional data, dimension reduction techniques such as projections are applied to map the input space to a 2D or 3D visual space. Many layout strategies have been proposed addressing different objectives that are targeted at distinct domains and applications. The resulting projected information is typically displayed in form of 2D scatter plots. The user’s perspective such as the role of visual attention and cognitive processing for a respective layout and task has not been addressed much. It is the goal of this work to investigate, how characteristics in the layout affect the cognitive process during task completion. Eye trackers are an effective means to capture visual attention over time. An eye tracker is used in a user study, where users have been asked to perform typical analysis tasks for projected multidimensional data such as relation seeking, behavior comparison, and pattern identification. Those tasks often involve detecting and correlating clusters. To understand the role of point density within clusters, cluster sizes, and cluster shapes, synthetic 2D scatter plots were created where properties could be manipulated manually. How changing various parameters affect the visual attention pattern has been investigated. The insight obtained from synthetic data is transferred to investigate the decision making with real-world data. Some conclusions on how different projection methods support or hinder decision-making leading to respective guidelines are drawn. In addition, this thesis shows that a 3D visual space can increase the performance of common visual analysis tasks due to a higher projection precision. The findings are backed up with a user study. However, 3D projections typically are displayed on a 2D screen which may impede the correct perception of the third dimension. This thesis presents a study that investigates the effect of stereoscopic environments when used for the visual analysis of multi-dimensional data after projection into a 3D visual space. Finally, through the last stage, visually encoding data clusters in a 3D setup as the form of enclosing surfaces or hulls are compared to scatter plots to evaluate the suitability of these methods for the visual analysis tasks. Efficient analysis of multi-dimensional data in order to understand the relationships between information hidden in huge data sets is essential. Our results offer interesting insight on the use of projection layouts in data visualization tasks and provide a departing point for further systematic investigations.