A Decreasing Binary Decision Tree Classification for Unbalanced Data in Customer Value Segmentation

摘要

The traditional data mining can not reflect the influence caused by the unbalanced quantity distribution of customers who have a variety of values.To solve the bias learning caused by unbalanced data in customer value segmentation,this paper proposes a classifying method based on decreasing binary decision tree,which transforms a multi-classification to a series of binary classifying trees.In every step of the method,the quantitative difference between the binary classes is decreased continually by removing the samples which are classified at the highest accuracy,and further classification only focused on the rest samples.The test result proves that this method can optimize the accuracy of all customer sample classes and produce effective rules for multi-classification in customer value segmentation.