曲线下的R逻辑回归区域

Question

我正在使用此页面执行逻辑回归.我的代码如下.

mydata <- read.csv("http://www.ats.ucla.edu/stat/data/binary.csv")
mylogit <- glm(admit ~ gre, data = mydata, family = "binomial")
summary(mylogit)
prob=predict(mylogit,type=c("response"))
mydata$prob=prob

运行此代码后,mydata dataframe有两列 - 'admit'和'prob'.这两列不应该足以获得ROC曲线吗？

如何获得ROC曲线.

其次,通过嘲笑mydata,似乎模型正在预测可能性admit=1.

那是对的吗？

如何找出模型预测的特定事件？

谢谢

更新:似乎以下三个命令非常有用.它们提供了最大精度的截止点,然后有助于获得ROC曲线.

coords(g, "best")

mydata$prediction=ifelse(prob>=0.3126844,1,0)

confusionMatrix(mydata$prediction,mydata$admit

Answer 1

ROC曲线比较预测和答案的等级.因此,您可以使用包评估ROC曲线,pROC如下所示:

mydata <- read.csv("http://www.ats.ucla.edu/stat/data/binary.csv")
mylogit <- glm(admit ~ gre, data = mydata, family = "binomial")
summary(mylogit)
prob=predict(mylogit,type=c("response"))
mydata$prob=prob
library(pROC)
g <- roc(admit ~ prob, data = mydata)
plot(g)    

Answer 2

另一种绘制ROC曲线的方法......

library(Deducer)
modelfit <- glm(formula=admit ~ gre + gpa, family=binomial(), data=mydata, na.action=na.omit)
rocplot(modelfit)

Answer 3

#Another way to plot ROC

mydata <- read.csv("http://www.ats.ucla.edu/stat/data/binary.csv")   
mylogit <- glm(admit ~ gre, data = mydata, family = "binomial")    
summary(mylogit)     
prob=predict(mylogit,type=c("response"))    
library("ROCR")    
pred <- prediction(prob, mydata$admit)    
perf <- performance(pred, measure = "tpr", x.measure = "fpr")     
plot(perf, col=rainbow(7), main="ROC curve Admissions", xlab="Specificity", 
     ylab="Sensitivity")    
abline(0, 1) #add a 45 degree line