Discovery Visualization and Visual Data Mining

View/ Open
Date
1999Author
Ribarsky, William
Katz, Jochen
Jiang, Frank
Holland, Aubrey
Metadata
Show full item recordAbstract
This paper describes discovery visualization, a new visual data mining approach that has as a key element the heightened awareness of the user by the machine. Discovery visualization promotes the concept of continuous interaction with constant feedback between man and machine and constant unfolding of the data. It does this by providing a combination of automated response and user selection to achieve and sustain animated action while the user explores time-dependent data. The process begins by automatically generating an overview using a fast clustering approach, where the clusters are then followed as time-dependent features. Discovery visualization is applied to both test data and real application data. The results show that the method is accurate and scalable, and it offers a straightforward, error-based process for improvement of accuracy.