2007 IEEE International Joint Conference on Neural Networks (IJCNN 2007), August 12-17, 2007, Orlando, Florida, USA
Visual data mining with virtual reality spaces are used for the representation of data and symbolic knowledge. The approach is illustrated with data from a geophysical prospecting case in which partially defined fuzzy classes are present. In order to understand the structure of both the data and knowledge extracted in the form of production rules, structure-preserving and maximally discriminative virtual spaces are constructed. High quality visual representationscan be obtained using Samman and Nonlinear Discriminant neural networks. Rough set techniques are used for demonstrating the irreducibility of the set of original attributes and for learning the symbolic knowledge. Grid computing techniques are used for constructing sets of virtual reality spaces andfor assessing the behavior of some of the neural network parameters controlling the quality of the virtual worlds. The general properties of the symbolic knowledge can be found with greater ease in the virtual reality space whereas both the prediction of unknown objects to the target class, as well as a derivation of a fuzzy membership function from the virtual reality space and the neural network results are obtained.
Proceedings of the 2007 IEEE International Joint Conference on Neural Networks (IJCNN 2007).