| Abstract | The information explosion requires the development of alternative data mining procedures that speed up the process of scientific discovery. The improved in-depth understanding and ease of interpretability of the internal structure of data by investigators allows focusing on the most important issues, which is crucial for the identification of valid, novel, potentially useful, and understandable patterns (regularities, oddities, surprises, etc). Computational visualization techniques are used to explore, in an immersive fashion, inherent data structure in both an unsupervised and supervised manner. Supervision is provided via i) domain knowledge contained in the data, and ii) unsupervised data mining procedures, such as fuzzy clustering, etc. The Virtual Reality (VR) approach for large heterogeneous, incomplete and imprecise (fuzzy) information is introduced for the problem of visualizing and analyzing general forms of data. The method is based on mappings between a heterogeneous space representing the data, and a homogeneous virtual reality space. This VR-based visual data mining technique allows the incorporation of the unmatched geometric capabilities of the human brain into the knowledge discovery process. Traditional means of interpretation would require more time and effort in order to achieve the same level of deep understanding of complex high dimensional data as the proposed technique. This hybrid approach has been applied successfully to a wide variety of real-world domains including astronomy, genomics, and geology, providing useful insights. |
|---|
