Improving of kNN Based on Attributes Bagging
Hanying Hu · Jisuanji gongcheng · 2005
This paper proposes the attributes Bagging to improve KNN. A simple alternative to bagging is to resampling attribute sets instead oftraining 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 BaggingkNN. Experiments with UCI datasets show the validity of attributes Bagging kNN.