数据流决策树集成分类算法综述

A SURVEY OF DECISION TREE ENSEMBLE CLASSIFICATION ALGORITHMS OF DATA STREAMS

数据流的集成分类方法可以提高预测精度或者可将复杂、困难的学习问题分解为更简单、容易的子问题,且对概念漂移有良好的适应性和恢复性。结合Bagging、Boosting和Stacking三种集成学习框架的工作原理和方法,分别对其相关决策树算法进行了分析和总结。详细介绍了数据流中的概念漂移问题,以及对不同类型的概念漂移的检测处理方法。从所属集成学习框架、对比算法、算法优缺点等多角度对数据流决策树集成分类算法进行了分析和总结。同时对数据流决策树集成分类算法的典型应用和主要平台作了详细介绍。对数据流集成分类领域中的研究趋势进行了探讨,并归纳出下一步的研究方向。

The ensemble classification method of data streams is an effective method that can improve the prediction accuracy or decompose complex and difficult learning problems into simpler and easier sub-problems,and has good adaptability and recovery for concept drift.Combining the working principles and methods of Bagging,Boosting and Stacking,the related decision tree algorithms were analyzed and summarized.The problem of concept drift in data streams and the detection and processing methods of different types of concept drift were introduced in detail.The decision tree ensemble classification algorithm of data streams was analyzed and summarized from the perspective of its own ensemble learning framework,comparison algorithm,advantages and disadvantages of the algorithm,etc.The typical application and main platform of decision tree ensemble classification algorithms of data streams were introduced in detail.The trend of research in the field of data streams ensemble classification was discussed and the next research direction was summarized.