基于变精度粗糙集的分类决策树构造方法

针对分类决策树构造时最优属性选择困难、难以适合大规模数据集的问题,提出新的属性选择标准--属性分类重要性测度,引入王信度和支持度,设计了基于变精度粗集理论的决策树算法.分类重要性测度可全面刻画属性的综合分类能力,且计算比信息增益简单.决策树生长过程中引入支持度和置信度,以控制决策树的生长,提高决策树对噪声数据集和不相容数据集的处理能力,减小决策树的规模.通过对UCI上5个不同规模和类型的数据集进行测试计算,结果表明算法效率高于ID3算法,与UCI报告的最好结果相当.

Study on constructing method of classifying decision tree based on variable precision rough set

Considering difficulty of choosing the best attribute and dealing with large-scale data set in constructing classifying decision tree,a new selection criterion called importance measure of attributes' classification(IMAC) and a decision tree constructing algorithm based on VPRS are proposed.The IMAC can describe classification capabilities of attributes comprehensively,and is simpler than traditional information entropy in calculation.In order to control growing up of the decision tree,confidence and support are introduced in algorithm;it can not only reduce the size of decision tree but also enhance the capability of decision tree in processing noise data and incompatible data.The proposed algorithm is tested with five different size and type of data sets in the UCI,the results show that proposed method is more efficient than ID3 algorithm,and equal to the best results of the UCI.