一种面向噪声数据的决策树优化算法研究

针对ID3算法构造的决策树结构复杂、对噪声数据比较敏感等局限性,提出一种新的面向噪声数据的决策树构造算法。算法借鉴变精度粗糙集和尺度函数概念,采用不同尺度下近似分类精度选择测试属性构造决策树,在算法形成过程中利用决策规则的可信度对决策树进行修剪,避免了生成的决策树过于庞大。结果表明,该方法是有效的,能够克服部分噪声数据对决策树的影响,且能满足不同用户对决策精度的要求。

Research of Decision Tree Optimization Algorithm for Noise Data

Aiming at the limitations of complex structure and lack of noise data adaptability for decision tree constructed by ID3,this paper proposes a new decision tree construction algorithm for noise data,which introduces the concept of variable precision rough set and scale function,uses the classification accuracy at different scales to select testing attributes and to build decision tree.The credibility of decision rule is put forward in the forming process of the algorithm to cut branches for decision tree,avoiding the problem of too large size of decision tree generated by the proposed algorithm.The results show that the method is effective and can overcome some of the impact of noise data,and can meet different users＇ requirements on the decision accuracy.