基于属性Bagging kNN性能的增强  

Improving of kNN Based on Attributes Bagging

作　　者：张震[1] 胡捍英[1] 

机构地区：[1]解放军信息工程大学信息工程学院,郑州450002

出　　处：《计算机工程》2005年第15期160-161,171,共3页Computer Engineering

基　　金：国防预研基金资助项目

摘　　要：提出了用于增强kNN的属性Bagging(ABagging),ABagging通过对属性重抽样而不是对训练实例重抽样来获得多个训练集。kNN对于属性重抽样不稳定,因而ABagging能有效降低kNN的错误率。ABaggingkNN对于不相关属性也有比kNN强得多的抵抗力。另外AbaggingkNN的速度也比BaggingkNN更快。用UCI数据集证明了ABaggingkNN的有效性。This paper proposes the attributes Bagging to improve KNN. A simple alternative to bagging is to resampling attribute sets instead of training sets. Attributes bagging kNN results in lower error rate than that of kNN because kNN is unstable when attributes bagging is employed.Furthermore attributes Bagging kNN is more robust to irrelevant attributes than kNN itself. Finally attributes bagging kNN is faster than Bagging kNN. Experiments with UCI datasets show the validity of attributes Bagging kNN.

关 键 词：BAGGING KNN 属性Bagging 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]