A study of method for knowledge discovery on set-valued features

A new method for classification on set-valued features is proposed and was used based on the adaptive subspace decomposition and separability index. In a high-dimensional original feature space, a few dimensions adapted for classification are selected from the subspaces. The classification rules are extracted from the decision information table based on the selected dimensions and binary relation about the set-valued features. An example of hyperspectral image classification was given, and an experimental investigation shows that it is an effective knowledge-based data fusion method.