一个机器学习定量属性的定性方法 TCIN  

A Method Transforming Continuous Attributes into Nominal Attributes

作　　者：毕建东 曲复宛[1,2] 

机构地区：[1]哈尔滨市电业局 [2]哈尔滨工业大学计算机科学与工程系应用软件教研室

出　　处：《哈尔滨工业大学学报》1997年第5期73-76,共4页Journal of Harbin Institute of Technology

摘　　要：一个机器学习定量属性的定性方法ＴＣＩＮ毕建东（计算机科学与工程系）曲复宛（哈尔滨市电业局）摘要提出了一个连续属性离散化方法ＴＣＩＮ，它首先使用自然划分法对区间进行划分，然后使用ＫＮ－近邻估计，利用基于最小错误率的Ｂａｙｅｓ决策寻找划分点进一步离散化连...In learning from examples, most of learning algorithms can only deal with nominal attributes Continuous valued attributes must bediscretised prior to induction Acontinuous valued attribute is typically discretized by partitioning its range into subranges This paper introduces a method TCIN which discretizes continuous valued attributes TCIN first divides some parts of interval which only contain attribute values of examples belonging to the same class Remaining parts are further partitioned by Bayesian decision and K N -neigbourhood estimate Finally a comparison between TCIN and MDLPC which is a discretised method of continuous attributes is made The correct rate (85%) of TCIN is higher than that (81%) o fMDLPC

关 键 词：示例学习 连续属性 离散化 TCIN 机器学习 

分 类 号：TP301[自动化与计算机技术—计算机系统结构]