Description

Data Mining may be defined as the (non-trivial) process of searching and analyzing data in order to find implicit but potentially useful information. It is build on the theories and technologies from many fields, including not only data mining and knowledge discovery, but also multivariate statistics, clustering analysis, multidimensional scaling, database interfaces, information retrieval, etc.

Human perceptual system processes different types of data in very flexible way. It automatically recognizes unusual properties, while at the same time ignores well-known properties. Therefore a human interaction with visualization is important, especially when a task is not strictly formulated. For instance, ``find ``strange'' customers'' or ``how can we attract more people to use our service''.

The continued hardware and software advances during the last few years make it possible to employ highly advanced visualisation techniques for the purpose of data mining. Specifically, it has become possible to do explorative data analysis in fully 3D immersive environments (or use advanced animations to simulate 3D immersive worlds).

The Ph.D. project addresses path finding methods that can be used for the purpose of 3D visual data mining. In 3D plots with a very large number of data points, the human eye does not easily recognize relationships and structures in the data set. The main goal of my Ph.D. work is to define, design, develop, and evaluate methods that facilitate the detection of relationships and structures in the data during navigation.

The goal of automatic path finding is to determine a path that the data analyst should follow when exploring the data. The task of finding a path can be divided into several tasks:

	finding interesting locations,
	calculating local paths for interesting locations,
	connecting local paths to a global path, and
	defining a view direction on a curve.

A general and widely used definition for an interesting location is a dense region, i.e., an accumulation (cluster) of points. I expect that dense regions also play an important role when navigating in immersive 3D worlds. In addition, it will be interesting to incorporate specific features of the data sets into the definition of interesting locations. Examples include the spreadness and symmetry of the data. I want to define and experimentally verify the notion of an interesting location for the purpose of explorative immersive data analyses.

In the case, when an interesting location in a 3-dimensional dataset forms a body the local path can be defined in terms of a (spiral) curve on a surface enclosed by the body. Following the spiral curve permits the data analyst to observe and analyze the body from different sides and angles.

Previously, we have noticed that the observer ``gets lost'' when no data points are visible. Therefore, it is important to consider the observer's view direction during navigation. A suitable view direction can be the direction to the center of an accumulation of points. Moreover, we experienced that sudden changes in the view direction and the lack of a gravity are problematic for the observer.

Further readings:

	A. Mazeika, M. Böhlen, P. Mylov. Density Surfaces for Immersive Explorative Data Analyses. In Proceedings of Workshop on Visual Data Mining in conjunction with SIGKDD, 2001 [.pdf]
	A. Mazeika, M. Böhlen, P. Mylov. Using Nested Surfaces to Detect Structures in Databases. In Proceedings of Workshop on Visual Data Mining, in conjunction with ECML/PKDD, 2001 [.pdf]