R语言实现K近邻(knn)算法

k近邻算法

需求

rm(list=ls())
knnProp=read.csv("kNN_fictious.csv",header = FALSE)
#样本个数
Prow=86
#测试数据的个数
Srow=258
#循环次数
forNum=30

#knn算法
knn <- function(P,S,K){
  #数组，存储最短的K个距离，样本分类标签和分类结果
  #前K列为距离，1+K到2K为样本分类标签
  #2K+1为分类结果，2K+2为原始分类标签
  #2K+3为分类结果是否正确的判断，T为正确，F为错误
  distanceArray=array(0,dim=c(Prow,2*K+3))
  for(i in 1:Prow){
    #临时数组，存储258个距离和样本分类标签
    distanceC=array(0,dim=c(Srow,2))
    for(j in 1:Srow){
      #计算欧式距离
      distance=sqrt((P[i,]$V1-S[j,]$V1)^2+
        (P[i,]$V2-S[j,]$V2)^2+
        (P[i,]$V3-S[j,]$V3)^2+
        (P[i,]$V4-S[j,]$V4)^2)
      #存储距离
      distanceC[j,1]=distance
      #存储样本分类标签
      distanceC[j,2]=S[j,]$V5
    }
    #距离排序
    distanceC=distanceC[order(distanceC[,1]),]
    #两个暂时变量，用于判断分类
    numA=0
    numB=0
    #存储前K个距离，样本分类标签
    for(k in 1:K){
      distanceArray[i,k]=distanceC[k,1]
      distanceArray[i,k+K]=distanceC[k,2]
      if(distanceC[k,2]==1){
        numA=numA+1
      } else{
        numB=numB+1
      }
    }
    #判断分类结果，并存储
    if(numA>numB){
      distanceArray[i,2*K+1]=1
    } else{
      distanceArray[i,2*K+1]=-1
    }
    #存储待分类数据的原始分类结果
    distanceArray[i,2*K+2]=P[i,]$V5
    #判断分类是否正确
    if(distanceArray[i,2*K+2]==distanceArray[i,2*K+1]){
      distanceArray[i,2*K+3]='T'
    } else{
      distanceArray[i,2*K+3]='F'
    }
  }
  return(distanceArray)
}

#求正确率
Accuracy <- function(distanceArray,K){
  #分类正确的个数
  TNum=0
  #分类错误的个数
  FNum=0
  #正确率判断
  for(i in 1:Prow){
    if(distanceArray[i,2*K+3]=='T'){
      TNum=TNum+1
    } else{
      FNum=FNum+1
    }
  }
  accuracy=TNum/(TNum+FNum)
  return(accuracy)
}

#随机抽取样本循环30次，求30次正确率
Sample <- function(prop,K){
  accuracy=c()
  for(i in 1:forNum){
    sampleNum=sample(1:344,344)
    distanceArray=knn(knnProp[sampleNum[1:86],],knnProp[sampleNum[87:344],],K)
    accuracy=c(accuracy,Accuracy(distanceArray,K))
  }
  rst=mean(accuracy)
  return(rst)
}
rst1=Sample(knnProp,3)
rst2=Sample(knnProp,7)
rst3=Sample(knnProp,11)
rst1
rst2
rst3