关键词: 数据流 排序二叉树 连续属性

Abstract:

Data stream mining classification is a very challenging job in the field of data mining. VFDT is a one-pass algorithm for decision tree construction.    It uses the Hoeffding inequality to achieve a probabilistie bound on the accuracy of the tree constructed. VFDTc improves VFDT, and make it be able to p  rocess continuous attributes. Based on VFDT and VFDTc , we design and realize an efficient algorithm VFDT-BSTree based on binary search trees. The algor  ithm solves the problems existing in VFDTc, and increases the speeds of dynamic sample insertion and best split node selection, and thus improves the sp eed of classification. The experimental results show that VFDT-BSTree＇s time is 32. 25% less than that of VFDT, and 24. 96% less than that of VFDTc on  average, while the same tree size and accuracy are kept.

Key words: data streams, binary search tree, continuous attribute