install.packages("tidyverse")
library(tidyverse)
install.packages("DataExplorer")
library(DataExplorer)
library(caret)
library(randomForest)
library(pROC)


rawdata <- read.csv(file.choose(), stringsAsFactors = F) # 打开data.csv
str(rawdata)

rawdata$gender <- as.factor(rawdata$gender)
rawdata$dyspnea <- as.factor(rawdata$dyspnea)
rawdata$class <- as.factor(rawdata$class)
rawdata$CD4 <- as.numeric(rawdata$CD4)


glimpse(rawdata)
plot_missing(rawdata)


rawdata3 <- as.data.frame(na.roughfix(rawdata))
plot_missing(rawdata3)


prop.table(table(rawdata3$class))
contrasts(rawdata3$class)

set.seed(9560)
up_rawdata3 <- upSample(x = rawdata3[, -ncol(rawdata3)],
                        y = rawdata3$class, 
                        yname = "class")                         
table(up_rawdata3$class)




set.seed(623)
head(up_rawdata3)
NCOL(up_rawdata3)
subsets = seq(2,16,1)
ctrl= rfeControl(functions = rfFuncs, method = "cv",verbose = FALSE, returnResamp = "final")
Profile = rfe(up_rawdata3[,-c(1,ncol(up_rawdata3))], up_rawdata3$class, sizes = subsets, rfeControl = ctrl)
print(Profile)
Profile$optVariables 
colnames(up_rawdata3)



indexFeature = which(colnames(up_rawdata3) %in% Profile$optVariables)  ## 这句是为了找出最优子集的在原来数据中列坐标
print(indexFeature)

head(up_rawdata3)
new_up_rawdata3 = up_rawdata3[,indexFeature]  

new_up_rawdata3$class = up_rawdata3$class

colnames(new_up_rawdata3)
str(new_up_rawdata3)




set.seed(42)
trains <- createDataPartition(y = new_up_rawdata3$class,
                              p = 0.8, 
                              list = F)
traindata <- new_up_rawdata3[trains, ]
testdata <- new_up_rawdata3[-trains, ]

prop.table(table(traindata$class))
prop.table(table(testdata$class))



tc_set <- trainControl(method = "repeatedcv", 
                       repeats = 5, 
                       number = 5,
                       classProbs = T,
                       summaryFunction = prSummary)
tune_set <- expand.grid(mtry = 1:6)

set.seed(42)

#install.packages('MLmetrics')
library("MLmetrics")

up_rfs <- train(x = traindata[,-ncol(traindata)],
                y = traindata$class,
                method = "rf",
                trControl = tc_set,
                tuneGrid = tune_set,
                importance = T)
up_rfs
plot(up_rfs)

final_rf <- up_rfs$finalModel


final_rf


par(mfrow=c(1,1))
plot(final_rf, main = "ERROR & TREES")
legend("top", 
       legend = colnames(final_rf$err.rate),
       lty = 1:3,
       col = 1:3,
       horiz = T)


importance(final_rf)
varImpPlot(final_rf, main = "varImpPlot", type=2)



partialPlot(x = final_rf, 
            pred.data = traindata,
            x.var = Myo,
            which.class = "death",
            ylab = "death")

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



trainpredprob <- predict(final_rf,
                         newdata = traindata,
                         type = "prob")

trainroc <- roc(response = traindata$class,
                predictor = trainpredprob[, 2]) 

trainroc$thresholds


plot(trainroc, 
     print.auc = TRUE, 
     auc.polygon = TRUE, 
     grid = c(0.1, 0.2),
     grid.col = c("green", "red"), 
     max.auc.polygon = TRUE,
     auc.polygon.col = "skyblue", 
     print.thres = TRUE)


trainpredlab <- as.factor(
  ifelse(trainpredprob[, 2] > 0.414, "survival", "death")
)

confusionMatrix(data = trainpredlab, 
                reference = traindata$class, 
                positive = "death",
                mode = "everything") 

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

testpredprob <- predict(final_rf,
                        newdata = testdata,
                        type = "prob")

testpredlab <- as.factor(
  ifelse(testpredprob[, 2] > 0.414, "survival", "death")
)

confusionMatrix(data = testpredlab,
                reference = testdata$class,
                positive = "death",
                mode = "everything")

testroc <- roc(response = testdata$class, 
               predictor = testpredprob[, 2]) 


plot(testroc, 
     print.auc = TRUE, 
     auc.polygon = TRUE, 
     grid = c(0.1, 0.2),
     grid.col = c("green", "red"), 
     max.auc.polygon = TRUE,
     auc.polygon.col = "skyblue", 
     print.thres = TRUE)








plot(trainroc, 
     print.auc = TRUE, 
     grid = c(0.1, 0.2),
     auc.polygon = F, 
     max.auc.polygon = T, 
     main = "模型效果图示",
     grid.col=c("green", "red"))
plot(testroc, 
     print.auc = TRUE, 
     print.auc.y = 0.4,
     add = T, 
     col = "red")
legend("bottomright", 
       legend = c("traindata", "testdata"),
       col = c(par("fg"), "red"), 
       lwd = 2, 
       cex = 0.9)

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


traindata <- cbind(traindata, trainpredprob)
traindata$predclass <- trainpredlab
write.csv(traindata, 
          "过抽样训练集数据加预测.csv", 
          row.names = F)

testdata <- cbind(testdata, testpredprob)
testdata$predclass <- testpredlab
write.csv(testdata, 
          "过抽样测试集数据加预测.csv", 
          row.names = F)