基于粒计算的决策树并行算法的应用  

Parallel decision tree algorithm based on granular computing

作　　者：周浩[1] 刘萍[1] 邱桃荣[1] 白小明[1] 

机构地区：[1]南昌大学计算机科学与技术系,江西南昌330031

出　　处：《计算机工程与设计》2015年第6期1504-1509,共6页Computer Engineering and Design

基　　金：国家自然科学基金项目(61070139);江西省科技支撑计划基金项目(20112BBG70087);江西省教育厅科技计划基金项目(GJJ14134;GJJ14143)

摘　　要：针对传统的决策树分类算法不能有效解决海量数据挖掘的问题,结合并行处理模型MapReduce,研究基于粒计算的ID3决策树分类的并行化处理方法。基于信息粒的二进制表示来构建属性的二进制信息粒向量,给出数据集的二进制信息粒关联矩阵表示;基于二进制信息粒关联矩阵,提出属性的信息增益的计算方法,设计基于MapReduce的粒计算决策树并行分类算法。通过使用标准数据集和实际气象领域的雷电真实数据集进行测试,验证了该算法的有效性。Because the traditional decision tree algorithm fails to solve the mass data mining ,combining with MapReduce ,the parallel ID3 algorithm based on the granular computing （GrC） was studied .Based on binary representation of information granu‐lar ,a binary vector of attribute was constructed ,a binary information granule correlation matrix of dataset was also given .On the basis of this ,a algorithm was proposed to compute information gain of attributes ,and a decision tree method using granular computing was also proposed ,which was a parallel classical algorithm based on MapReduce .UCI benchmark datasets and the real thunder data from meteorological bureau were used in the experiments to verify the effectiveness the presented algorithm .

关 键 词：粒计算 并行 映射规约 决策树 信息增益 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]