Xgboostと他の手法比較.R

##様々な手法で予測精度の比較を行う##
##randomForest,SVM,ナイーブベイズ，ニューラルネット
library(xgboost)
library(caret)
library(Matrix)
library(kernlab)
library(randomForest)
library(e1071)
library(nnet)
library(ggplot2)
titanic<-read.csv("タイタニック乗船者リスト.csv")##データの読み込み
titanic<-na.omit(titanic)##欠損値削除
titanic$Survived<-as.factor(titanic$Survived)##質的変数に変換
sim<-10##シミュレーション回数
result<-matrix(0,sim,5)##結果の箱

for(i in 1:sim){
  ####学習データとテストデータに分ける####
  train.id<-sample(nrow(titanic),400)
  train.data<-titanic[train.id,-1]
  test.data<-titanic[-train.id,-1]
  
  
  ##randomForest##
  rf<-randomForest(Survived~.,train.data)
  test.data.rf<-cbind(test.data,"predict"=predict(rf,test.data))
  result[i,1]<-sum(test.data.rf$predict==test.data.rf$Survived)/nrow(test.data.rf)


  ##SVM##
  svm<-ksvm(Survived~.,train.data)
  test.data.svm<-cbind(test.data,"predict"=predict(svm,test.data))
  result[i,2]<-sum(test.data.svm$predict==test.data.svm$Survived)/nrow(test.data.svm)

  ##ナイーブベイズ##
  nb<-naiveBayes(Survived~.,train.data)
  test.data.nb<-cbind(test.data,"predict"=predict(nb,test.data))
  result[i,3]<-sum(test.data.nb$predict==test.data.nb$Survived)/nrow(test.data.nb)

  ##ニューラルネット##
  nn<-nnet(Survived~.,train.data,size=5)
  test.data.nn<-cbind(test.data,"predict"=round(predict(nn,test.data),0))
  result[i,4]<-sum(test.data.nn$predict==test.data.nn$Survived)/nrow(test.data.nn)

  ##XGboost##
  label.data.train<-as.integer(train.data$Survived)-1##目的変数を01に
  label.data.predict<-as.integer(test.data$Survived)-1
  
  train.data.xg<-train.data[,-1]##目的変数取り除く
  test.data.xg<-test.data[,-1]
  
  xgb.data<-xgb.DMatrix(data.matrix(train.data.xg),label=label.data.train)##Xgb型へ成型
  xgb.data.predict<-xgb.DMatrix(data.matrix(test.data.xg))
  
  param <- list(
    "objective" = "binary:logistic"
    ,"eta" = 0.01
    # ,"max_depth" = 10
    ,"min_child_weight" = 5
    # ,"subsample" = 1
    # ,"colsample_bytree" = 1
  )##パラメータ適当
  
  model <- xgboost(
    param=param
    ,data=xgb.data
    ,nrounds=1000
  )##実際の学習

  predict_xgb<-predict(model,xgb.data.predict)
  
  a<-c()
  m<-length(predict_xgb)#行数
  
  for(l in 1:m){
    if(predict_xgb[l]>0.5){a[l]<-1}
    else{a[l]<-0}
  }
  result[i,5]<-sum(a==label.data.predict)/length(a)
  
}

result.c<-apply(result,2,mean)##simの平均値
result.c<-t(as.matrix(result.c))
result.c<-rbind(result.c,result.c)
##グラフ描画
colnames(result.c) <- c("RF","SVM","Bayes","NN","xgb")
df<-data.frame(method=colnames(result.c),result=colMeans(result.c))
gp<-ggplot(df,aes(x=method,y=result,fill=method))
gp<-gp+geom_bar(width=0.8,stat="identity")
gp<-gp+ylab("result")
gp<-gp+geom_text(label=colMeans(result.c),vjust=1)
plot(gp)