一个改进的决策树算法

Iterative Dichotomiser version3（ID3）算法是数据挖掘中经典的决策树分类算法,其核心是分裂训练集属性的选择标准,即分裂前后的信息增益量最大,用该标准选择属性时对于取值较多的属性具有较强依赖性。剖析了ID3算法存在的不足并加以改进,引入了属性关注度,提出了改进算法AAID3算法。实验表明改进算法对原ID3算法的取值偏向问题有所克服并使分类更加准确,决策树更加简明。

An Enhanced ID3 Algorithm Based on Attribute Attention

In decision tree sorting,Iterative Dichotomiser version 3（ID3） is a classical algorithm used to generate a decision tree invented by Ross Quinlan.Its core lies in the split-off training which lumped the standard of attributes preference,namely,information gain is of maximum quantity both before split-off and after split-off.The use of this preference standard to select attributes has a strong dependence in regard of the milti-valued attributes preference.Disadvantages of ID3 algorithm were analyzed and improved through introducing attribute attention.To the effect,AAID3 algorithm was proposed.Experimental results expatiated AAID3 algorithm is superior to the ID3 in accuracy of classification,concision of decision tree,and independence from multi-valued attributes.