Exploring multivariate data through the application of visual analytics
Abstract
As data sources become larger and more complex, the ability to effectively explore and analyze patterns amongst varying sources becomes a critical bottleneck in analytic reasoning. Incoming data contains multiple variables, high signal to noise ratio, and a degree of uncertainty, all of which hinder exploration, hypothesis testing, and decision making. This work presents a suite of analytical methods coupled with novel visualization modalities in an interactive framework, that enables analysts and decision makers to explore their data in a visual analytics environment. This research couples traditional multivariate exploration methods, such as clustering, density estimation and time series modeling, into a user-centered interactive visual environment and demonstrates their efficacy. From the exploration of categorical, geospatial multivariate data to more traditional scientific multivariate data sources, the tools presented here aid the analyst in the generation and exploration of data hypotheses. This work presents a suite of visual analytics methods introduced into various multivariate data exploration systems. Methods include the use of linked views and interactive filtering for data exploration in spatiotemporal data, the application of kernel density estimation for data visualization in any arbitrary spatial domain, and the use of k-means clustering in data space for visualization in a geographic domain. Furthermore, data modeling in both space and time has been utilized with a focus on non-parametric modeling using seasonal trend decomposition with loess smoothing. Finally, the application of these various modeling tools for synthetic data creation within a visual analytics framework has also been accomplished.
Degree
Ph.D.
Advisors
Ebert, Purdue University.
Subject Area
Computer Engineering
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.