titanic_分类_泰坦尼克_随机森林_

library(ggplot2) library(psych) library(nnet) train=read.csv("C:/Users/秦文昭/Documents/python/titanic/train.csv",header=T,stringsAsFactors = FALSE) test=read.csv("C:/Users/秦文昭/Documents/python/titanic/test.csv",header=T,stringsAsFactors = FALSE) #str(train) #str(test) test=data.frame(test,"Survived"=NA) test=test[,c(1,12,2,3,4,5,6,7,8,9,10,11)] alldata=rbind(train,test) #===================================数据的导入与整理============================ #pclass #windows() #Pclass_S <- table(train$Survived, train$Pclass) #Pclass_S_prop <- prop.table(Pclass_S, 2) #ggplot(data = train, aes(x = Pclass, fill = factor(Survived)))+geom_bar(stat='count', position='dodge') + scale_x_continuous(breaks=c(1:3)) + labs(x = 'Pclass') #sex #windows() #Sex_S <- table(train$Survived, train$Sex) #Sex_S_prop <- prop.table(Sex_S, 2) #ggplot(data = train, aes(x = Sex, fill = factor(Survived)))+geom_bar(stat='count', position='dodge') #age #windows() #Agedata <- as.numeric(unlist(train$Age)) #Age_S <- table(train$Survived, cut(Agedata, breaks = c(0, 15, 30, 45, 60, 75, 90), labels = c('kids', 'teenagers', 'prime', 'middle', 'agedness', 'senium' ))) #Age_S_prop <- prop.table(Age_S, 2) #ggplot(data = data.frame(train$Survived, Agedata), aes(x = cut(Agedata, breaks = c(0, 15, 30, 45, 60, 75, 90)), fill = factor(train$Survived)))+geom_bar(stat='count', position='dodge') + labs(x = 'Age') + scale_x_discrete(labels = c('kids', 'teenagers', 'prime', 'middle', 'agedness', 'senium')) #family #Sibsp_S <- table(train$Survived, train$SibSp) #Parch_S <- table(train$Survived, train$Parch) #Sibsp_S_prop <- prop.table(Sibsp_S, 2) #Parch_S_prop <- prop.table(Parch_S, 2) #windows() #ggplot(data = train, aes(x = SibSp, fill = factor(Survived)))+geom_bar(stat='count', position='dodge') + scale_x_continuous(breaks=c(0:8)) + labs(x = 'Sibsp') #ggplot(data = train, aes(x = Parch, fill = factor(Survived)))+geom_bar(stat='count', position='dodge') + scale_x_continuous(breaks=c(0:6)) + labs(x = 'Parch') Families <- train$SibSp +train$Parch #ggplot(data = train, aes(x = Families, fill = factor(Survived)))+geom_bar(stat='count', position='dodge') + scale_x_continuous(breaks=c(0:10)) + labs(x = 'Families') family_label=seq(1:nrow(train)) for (i in 1:nrow(train)) { if(Families[i]<=3 & Families[i]>=1){family_label[i]=1} if((Families[i]>3 & Families[i]<=6) | (Families[i]==0)){family_label[i]=2} if(Families[i]>6){family_label[i]=3} } train=data.frame(train,'family_label'=family_label) #windows() #fl_S <- table(train$Survived, train$family_label) #fl_S_prop <- prop.table(fl_S, 2) #ggplot(data = train, aes(x = family_label, fill = factor(Survived)))+geom_bar(stat='count', position='dodge') + scale_x_continuous(breaks=c(1:3)) + labs(x = 'family_label') Families <- alldata$SibSp + alldata$Parch family_label=seq(1:nrow(alldata)) for (i in 1:nrow(alldata)) { if(Families[i]<=3 & Families[i]>=1){family_label[i]=1} if((Families[i]>3 & Families[i]<=6) | (Families[i]==0)){family_label[i]=2} if(Families[i]>6){family_label[i]=3} } alldata=data.frame(alldata,'family_label'=family_label) #fare #windows() #plot(density(train[train$Survived==0,]$Fare),main='fare',col='red') #lines(density(train[train$Survived==1,]$Fare),col='blue') #legend('topright',legend=c('Survived=0','Survived=1'),col=c("red","blue"),lty=1,lwd=2) #carbin #缺失值填充初始化 train[train$Cabin=="",]$Cabin="U" alldata[alldata$Cabin=="",]$Cabin="U" split_Initials<-function(df){ tmp=seq(1:nrow(df)) for(i in 1:nrow(df)){ tmp[i]=unlist(strsplit(df$Cabin[i],""))[1] } return(tmp) } deck=split_Initials(train) train=data.frame(train,'deck'=deck) deck=split_Initials(alldata) alldata=data.frame(alldata,'deck'=deck) #windows() #deck_S <- table(train$Survived, train$deck) #deck_S_prop <- prop.table(deck_S, 2) #ggplot(data = train, aes(x = deck, fill = factor(Survived)))+geom_bar(stat='count', position='dodge') #name title_list=list(c(' Capt', ' Col', ' Major', ' Dr', ' Rev'), c(' Don', ' Sir', ' the Countess', ' Dona', ' Lady'), c(' Mme', ' Ms', ' Mrs'),c(' Mlle', ' Miss'), " Mr", c(' Master',' Jonkheer')) names(title_list) <- c("Officer", "Royalty", "Mrs", "Miss", "Mr", "Master") key_map<-function(list=title_list,ele){ for(i in 1:length(list)){ if(ele %in% list[[i]]){map_name=names(list)[i]} } return(map_name) } split_name<-function(df){ surname=seq(1:nrow(df)) title=seq(1:nrow(df)) for(i in 1:nrow(df)){ tmp=unlist(strsplit(df$Name[i],",")) surname[i]=tmp[1] tmp1=tmp[2] title[i]=unlist(strsplit(tmp1,"[.]"))[1] title[i]=key_map(ele=title[i]) } return(data.frame("surname"=surname,"title"=title)) } tmp=split_name(train) train=data.frame(train,tmp) tmp=split_name(alldata) alldata=data.frame(alldata,tmp) #windows() #title_S <- table(train$Survived, train$title) #title_S_prop <- prop.table(title_S, 2) #ggplot(data = train, aes(x = title, fill = factor(Survived)))+geom_bar(stat='count', position='dodge') #embark #windows() #Embarked_S <- table(train$Survived, train$Embarked) #Embarked_S_prop <- prop.table(Embarked_S, 2) #ggplot(data = train, aes(x = Embarked, fill = factor(Survived)))+geom_bar(stat='count', position='dodge') #====================================特征工程=============================== #缺失值处理 #alldata_cor=alldata[,-c(4,7,8,9,11,15)] #windows() #pairs.panels(alldata_cor[,-c(1,2)])#可视化 #embark alldata[alldata$Embarked=="",]$Embarked="C" #fare tmp=alldata[alldata$Pclass==3,] alldata[is.na(alldata$Fare),]$Fare=median(tmp[tmp$Embarked=="S",]$Fare,na.rm=T) #alldata[is.na(alldata$Fare),] #age alldata_age=read.csv("C:/Users/秦文昭/Documents/python/titanic/alldata_age.csv",header=T,stringsAsFactors = FALSE) alldata$Age=alldata_age$Age #sum(is.na(alldata$Age)) rm(alldata_age) rm(tmp) #======================================model============================= library(foreach) library(caret)#这两个是做交叉验证用的 library(pROC) alldata=alldata[,-c(4,7,8,9,11,15)] alldata[,c(5,6)]=scale(alldata[,c(5,6)],scale=F)#2个连续型变量做归一化处理 #哑变量 class_sex=class.ind(alldata$Sex) class_embarked=class.ind(alldata$Embarked) names(class_embarked)<-c("embarked_C","embarked_Q","embarked_S") class_family=class.ind(alldata$family_label) names(class_family)<-c("fl1","fl2","fl3") class_deck=class.ind(alldata$deck) class_title=class.ind(alldata$title) alldata=data.frame(alldata[,c(1,2,3,5,6,8)],class_sex,class_embarked,class_family, class_deck,class_title) #r哑变量 id=alldata[,1] alldata=alldata[,-1] test=alldata[is.na(alldata$Survived),] train=alldata[1:891,] seperate<-function(y,alpha=0.5){#如果有模型求出来是概率，就用这个分 for (i in 1:length(y)) { if(y[i]<0.5){y[i]=0 }else{y[i]=1} } return(y) } #========================================knn============================== #划分数据集 set.seed(666) index<-createDataPartition(y=train$Survived,p=0.8,list=FALSE) survived_train<-train[index,] survived_test<-train[-index,] library(class) accuracy <- vector() for(i in 1:30){ KNN_M <- knn(train = survived_train[, -1], test = survived_test[,-1], cl = survived_train[, 1], k = i) CT <- table(survived_test$Survived, KNN_M) accuracy <- c(accuracy, sum(diag(CT))/sum(CT)*100) } max(accuracy) which.max(accuracy) set.seed(123) random <- sample(1:891, 104) result_KNN_R <- train[random, 1] accuracy_KNN_R <- vector() for(i in 1:30){ KNN_M <- knn(train = train[-random, -1], test = train[random, -1], cl = train[-random, 1], k = i) CT_KNN_R <- table(result_KNN_R, KNN_M) accuracy_KNN_R <- c(accuracy_KNN_R, sum(diag(CT_KNN_R))/sum(CT_KNN_R)*100) } accuracy_KNN_R_Max <- max(accuracy_KNN_R) accuracy_KNN_R_Max which.max(accuracy_KNN_R) #=================================