An experiment in machine learning of redundant knowledge
Igor Kononenko · 2002
Experiments in generating redundant diagnostic rules from examples in three medical domains are described. The idea is to generate a number of sets of decision rules (theories) using known inductive learning techniques. Each set is applied when classifying new objects. An object is classified to the class that is preferred by the majority of theories. The redundant knowledge with voting principle significantly outperformed the one theory principle. In addition, redundant knowledge generated in this way provides the possibility of better explanations, which is one of weak points of the inductively generated (nonredundant) sets of decision rules.>