Titanic_Classification

library(tidyverse) library(dplyr) library(caret) library(ModelMetrics) library(randomForest) library(stringr) library(xgboost) # for fitting GBMs library(ranger) # for fitting random forests library(rpart) library(patchwork) library(DataExplorer) library(vip)

rm(list=ls())

getwd()

setwd("D:/Non_Documents/AI/R/data/220810_181047_aifactory") dir()

train <- read.csv("train.csv", stringsAsFactors=T, na.strings=c("-999", "NA", ""))

test <- read.csv("test.csv", stringsAsFactors=T, na.strings=c("-999", "NA", "")) submission <- read.csv("sample_submission.csv", stringsAsFactors=T)

############

survival - 생존유무, target 값. (0 = 사망, 1 = 생존)

pclass - 티켓 클래스. (1 = 1st, 2 = 2nd, 3 = 3rd)

sex - 성별

Age - 나이(세)

sibsp - 함께 탑승한 형제자매, 배우자 수 총합

parch - 함께 탑승한 부모, 자녀 수 총합

ticket - 티켓 넘버

fare - 탑승 요금

cabin - 객실 넘버

embarked - 탑승 항구기

head(train) head(test) head(submission)

na 가 있는 열 확인하기

colSums(is.na(train)) ## Age 177개 결측치 colSums(is.na(test)) ## Age 86개, Fare 1개 결측치

결측치 대체 (KNN)

library(DMwR2) train2 <- knnImputation(train ,k=10) #KNN 대체법 실시 test2 <- knnImputation(test ,k=10) #KNN 대체법 실시 colSums(is.na(train2)) colSums(is.na(test2)) ## Age 86개, Fare 1개 결측치

str(train2) str(test2)

id 열 제거하기

train2 <- train2[,-c(1,4,9,11)] test2 <- test2[,-c(1,3, 8, 10)]

target -> factor로 바꾸기

train2$Survived <- as.factor(train2$Survived)

str(train2) str(test2)

pairs(train2) ## 간단히 상관관계 보기

############### GGally 패키지를 이용한 EDA ##############################

library(ggplot2) library(GGally)

#create pairs plot ggpairs(train_s)

ggpairs(train_s, mapping = aes(color = continent, alpha=0.3))

###################################### EDA ##################################

####DataExplorer 패키지를 이용한 EDA

introduce(train) plot_intro(train) ## 기본 plot_missing(train) ## NA 유무 plot_bar(train) ## 범주형 변수 시각화화

plot_histogram(train, ncol = 3) ## 연속형 변수 시각화 plot_density(train, ncol = 3) plot_qq(train, ncol = 3) ## 연속형 변수 정규성 plot_qq(train, by = "contract_until") ## 그룹별로 볼 수 있음

plot_correlation(train, type = "continuous") ##상관행렬(공분산)

plot_boxplot(train[c("value", "contract_until")], by = "contract_until")

plot_prcomp(train, ggtheme = theme_bw(), variance_cap = 0.50) ## 주성분 분석

더미변수 만들기 (범주형 -> 연속형)

#dummify(train["wday"]) %>% head()

create_report(data = train, y="value") ## 빠른 종합 레포트

############### WVplots 패키지를 이용한 EDA ##############################

library(WVPlots) train_s <- train %>% sample_n(100) %>% as_tibble()

PairPlot(train_s, colnames(train_s)[1:10], title = "aaa", group_var = NULL, palette="Dark2",point_color = "darkgray")

##################################################################################

#aov 아노바 분석 (독립변수)

#install.packages("multcomp") library(multcomp) library(ggplot2) cholesterol summary(cholesterol) aggregate(response ~ trt, data = cholesterol, mean) ggplot(cholesterol, aes(x = trt, y = response)) + geom_boxplot(notch = T) result <- aov(response ~ trt, data = cholesterol) anova(result) plot(result, which=1:3) shapiro.test(result$residuals) bartlett.test(response ~ trt, data = cholesterol) TukeyHSD(result) par(las=2, mar=c(5, 8, 4, 2)) plot(TukeyHSD(result))

clipr::write_clip(train)

str(train)

Simple interaction plot

interaction.plot(x.factor = train$continent, trace.factor = train$prefer_foot, response = train$value, fun = mean, type="b", col=c("black","red","green"), ### Colors for levels of trace var. pch=c(19, 17, 15), ### Symbols for levels of trace var. fixed=TRUE, ### Order by factor order in data leg.bty = "o")

#PCA 하기

train_dt <- prcomp(train, center = T,scale. = T) #데이터 표준화 포함

#PCA 결과 확인 train_dt

#PCA 결과 시각화

Proportion of variance 출력

plot(pca_dt,type = "l") screeplot(pca_dt, main = "", col = "green", type = "lines", pch = 1, npcs = length(pca_dt$sdev))

summary(pca_dt)

###################################### 회귀분석 ###############################

#훈련/검증 데이터 70:30 으로 나누기

idx<-createDataPartition(train2$Survived , p=0.7, list=F) train_df<-train2[idx,] test_df<-train2[-idx,]

#모델 만들기 m1<-train(Survived~., data=train_df, method="glm") #로지스틱 회귀 모델

m2<-randomForest(Survived~., data=train_df, ntree=100) #랜덤포레스트 모델

Fit a single regression tree

tree <- rpart(Survived ~ ., data = train_df)

Fit a random forest

set.seed(101) rfo <- ranger(Survived ~ ., data = train_df, importance = "impurity")

Fit a GBM

set.seed(102) bst <- xgboost( data = data.matrix(subset(train_df, select = -Survived)), label = train_df$Survived, objective = "reg:linear", nrounds = 100, max_depth = 5, eta = 0.3, verbose = 0 # suppress printing )

VI plot for single regression tree

vi_tree <- tree$variable.importance barplot(vi_tree, horiz = TRUE, las = 1)

VI plot for RF

vi_rfo <- rfo$variable.importance %>% sort() barplot(vi_rfo, horiz = TRUE, las = 1)

VI plot for GMB

library(Ckmeans.1d.dp) vi_bst <- xgb.importance(model = bst) xgb.ggplot.importance(vi_bst)

i1 <- vip(m1) + ggtitle("Logistic regression") i2 <- vip(m2)+ ggtitle("Random Forest") i3 <- vip(tree)+ ggtitle("Descision tree") i4 <- vip(rfo)+ggtitle("Fast Random Forest") i5 <- vip(bst)+ggtitle("XGBoost")

i3+i1+i5+i2+i4

#예측하기

p1<-predict(m1, test_df) p2<-predict(m2, test_df) p3<-predict(tree, test_df) p3 <- as.factor(round(p3[,2])) p4<- predict(rfo, data = test_df, predict.all = TRUE) p4 <- p4$predictions[,2] p4 <- as.factor(ifelse(p4==1, 0, 1)) p5<-predict(bst, data.matrix(test_df[,-1])) p5 <- as.factor(ifelse(round(p5)==1, 0, 1))

#평가하기 (Accuracy)

r1 <- caret::confusionMatrix(test_df$Survived, p1)$overall[1] #로지스틱 회귀분석 r2 <- caret::confusionMatrix(test_df$Survived, p2)$overall[1] #랜덤포레스트 r3 <- caret::confusionMatrix(test_df$Survived, p3)$overall[1] #의사결정나무 r4 <- caret::confusionMatrix(test_df$Survived, p4)$overall[1] #ranger r5 <- caret::confusionMatrix(test_df$Survived, p5)$overall[1] #xgboost

name <- c("Logistic regression", "Random Forest", "Descision tree", "Fast Random Forest","XGBoost") r_accuracy <- round(c(r1,r2,r3,r4,r5),2) v <- as.data.frame(cbind(name, r_accuracy) )

v %>% mutate(name = fct_reorder(name,desc(r_accuracy))) %>% ggplot(aes(name, r_accuracy, fill=name))+geom_col() + geom_text(data = v, aes(label = paste("Accuracy=",r_accuracy)), y = r_accuracy, size=5)+ ggtitle("Titanic 생존자 예측")+ labs(x="ML Model", y="Acuuracy", subtitle="")+ theme_bw()+ theme(axis.text.y = element_text(size=12), axis.text.x = element_text(size=12))+ theme(legend.position="none")

#평가하기 (F1)

f1 <- caret::confusionMatrix(test_df$Survived, p1)$byClass[7] #로지스틱 회귀분석 f2 <- caret::confusionMatrix(test_df$Survived, p2)$byClass[7] #랜덤포레스트 f3 <- caret::confusionMatrix(test_df$Survived, p3)$byClass[7] #의사결정나무 f4 <- caret::confusionMatrix(test_df$Survived, p4)$byClass[7] #ranger f5 <- caret::confusionMatrix(test_df$Survived, p5)$byClass[7] #xgboost

name_f1 <- c("Logistic regression", "Random Forest", "Descision tree", "Fast Random Forest","XGBoost") r_f1 <- round(c(f1,f2,f3,f4,f5),2) v_f1 <- as.data.frame(cbind(name, r_f1) )

v_f1 %>% mutate(name_f1 = fct_reorder(name,desc(r_f1))) %>% ggplot(aes(name_f1, r_f1, fill=name_f1))+geom_col() + geom_text(data = v_f1, aes(label = paste("F1=",r_f1)), y = r_f1-0.2, size=5)+ ggtitle("Titanic 생존자 예측")+ labs(x="ML Model", y="F1", subtitle="")+ theme_bw()+ theme(axis.text.y = element_text(size=12), axis.text.x = element_text(size=12))+ theme(legend.position="none")

#평가하기 (ROC, AUC) library(pROC)

roc1 <- pROC::roc(as.numeric(test_df$Survived), as.numeric(p1)) #로지스틱 회귀분석 roc2 <- pROC::roc(as.numeric(test_df$Survived), as.numeric(p2)) #랜덤포레스트 roc3 <- pROC::roc(as.numeric(test_df$Survived), as.numeric(p3)) #의사결정나무 roc4 <- pROC::roc(as.numeric(test_df$Survived), as.numeric(p4)) #ranger roc5 <- pROC::roc(as.numeric(test_df$Survived), as.numeric(p5)) #xgboost

name_roc <- c("Logistic regression", "Random Forest", "Descision tree", "Fast Random Forest","XGBoost") r_roc <- round(c(roc1$auc,roc2$auc, roc3$auc, roc4$auc, roc5$auc),2) v_roc <- as.data.frame(cbind(name, r_f1) )

v_roc %>% mutate(name_roc = fct_reorder(name_roc,desc(r_roc))) %>% ggplot(aes(name_roc, r_roc, fill=name_roc))+geom_col() + geom_text(data = v_roc, aes(label = paste("AUC=",r_roc)), y = r_roc-0.1, size=5)+ ggtitle("Titanic 생존자 예측")+ labs(x="ML Model", y="AUC", subtitle="")+ theme_bw()+ theme(axis.text.y = element_text(size=12), axis.text.x = element_text(size=12))+ theme(legend.position="none")

dev.off() par(mfrow=c(1,5)) roc_p1 <- plot.roc(roc1, legacy.axes = T, main="Logistic regression") roc_p2 <- plot.roc(roc2, legacy.axes = T, main="Random Forest") roc_p3 <- plot.roc(roc3, legacy.axes = T, main="Descision tree") roc_p4 <- plot.roc(roc4, legacy.axes = T, main="Fast Random Forest") roc_p5 <- plot.roc(roc5, legacy.axes = T, main="XGBoost")

최종 랜덤포레스트 모델로 최종 모델링 하기

m<-randomForest(Survived~., data=train2, ntree=100)

p<-predict(m, test2)

############ Importance 시각화 ################

varImpPlot(m, main="varImpPlot of iris")

library(vip)

vi(m) vip(m)

#데이터 제출

p값을 문자열로 바꾸고 csv 파일로 제출하기

p<-as.character(p) submission$Survived <- p

write.csv(submission, "submission.csv", row.names=F)