Complexity-driven Evolution of Decision Graphs for Classification of Medical Data. Informatica 29:41–51

In the paper we study the possibility of constructing decision graphs with the help of several meta agents. Decision graphs are an extension of the well known decision trees and introduce the possibility of program nodes and cycles in a classification model. A two-leveled evolutionary algorithm for the induction of decision graphs is presented and the principle of classification based on the decision graphs is described. Several agents are used to construct the decision graphs; they are constructed and evolved with the help of automatic programming and evaluated with a universal complexity measure. The developed model is applied to a medical dataset for the classification of patients with mitral valve prolapse syndrome.