Interpreting fuzzy clustering results with virtual reality-based visual data mining: Application to microarray gene expression data

This paper combines fuzzy clustering with a virtual reality based technique for visual data mining. The purpose is to construct virtual reality spaces preserving as much structural information from the original data as possible, where the results of fuzzy clustering procedures can be displayed and analyzed. The construction of such spaces involves non-linear transformations of the original feature space, which can be either the space of the original attributes or the space of the fuzzy memberships with respect to the constructed fuzzy classes. In particular, the representation involves the centroids of the different classes, the individual memberships of all of the studied objects with respect to all of the fuzzy classes, and eventually their comparison with additional crisp partitions or partitions induced by a decision attribute. This approach is applied to different data sets from the fields of biology and medicine, including microarray gene expression data related to Alzheimer´s disease and Leukemia. The visual inspection and the navigation in the virtual reality spaces, provided useful insights about (i) the quality of the obtained classifications, (ii) the overlapping of different classes, and (iii) their relationships.