A New Classifiction Method to Overcome Over-Branching

Classification is an important technique indata mining. The decision trees built by most of the existingclassification algorithms commonly feature over-branching, which willlead to poor efficiency in the subsequent classification period. Inthis paper, we present a new value-oriented classification method,which aims at building accurately proper-sized decision trees whilereducing over-branching as much as possible, based on the concepts offrequent-pattern-node and exceptive-child-node. The experiments showthat while using relevant analysis as pre-processing, ourclassification method, without loss of accuracy, can eliminate theover-branching greatly in decision trees more effectively andefficiently than other algorithms do.