Classifier Building by Reduction of an Ensemble of Decision Trees to a Set of Rules

The paper presents a new approach for building classifiers by transforming an ensemble of classifiers into a single rule set. The proposed method improves generalization abilities of an ensemble with additional reduction of its complexity. It is dedicated to committees of decision trees and bases on transformation of a set of trees into a set of rules with a new, well-suited, weighed voting algorithm. The paper also presents experiments showing the properties and effectiveness of proposed method and direction of further research.