Pruning and rule extraction using class entropy

The described methodology addresses domain-characterization and knowledge-explicitation in difficult decision-making problems. In the framework of best exploiting class information, several pruning methods lead to effective tools for analyzing the complexity of a representation problem. A divide-and-conquer strategy results in a hierarchical procedure for extracting rules to synthesize the observed domain. Application to a real, complex clinical problem provides a valuable and satisfactory testbed for all aspects of the described methodology