适应类别增量的决策树训练算法

Adaptive Algorithm for Class Incremental Induction of Decision Tree

对于模式经常发生变化的客户资信评估、垃圾邮件检测和网络入侵检测等在线分类系统来说,自动感知客观存在的新类别,并让系统中的分类器对此作出自适应调整是其正确持续运行必须解决的问题。该文提出了一种适应新类别增加的决策树训练算法,该算法在新类别已检出的前提下,在原有决策树基础上利用新类别样本增量训练出新的决策树。实验结果表明:该文提出的算法可以较好地解决该问题,而与重新训练新决策树相比,它在分类器离线调整上较少的时间花费使其适用于在线分类系统。

In online classification system, such as credit evaluation, spare detection and intrusion detection, it is an important problem to detect the new pattern, and make the classifier to adapt for emerging pattern. Aiming at this problem, this paper proposes an incremental algorithm to train decision tree based on original computation and samples with new class. The result of experiment shows that the proposed algorithm can deal with incremental class and suit online classification system because of its little time cost.