ROC Curve and AUC

[안재형]

> library(MASS)
> data(cats)
> cats[1:10,]
   Sex Bwt Hwt
1    F 2.0 7.0
2    F 2.0 7.4
3    F 2.0 9.5
4    F 2.1 7.2
5    F 2.1 7.3
6    F 2.1 7.6
7    F 2.1 8.1
8    F 2.1 8.2
9    F 2.1 8.3
10   F 2.1 8.5

### Sex ~ Bwt and Sex ~ Hwt

# AUC

> install.packages("caTools")

> library(caTools)

> colAUC(cats[,2:3],cats[,1],plotROC=TRUE)
              Bwt      Hwt
F vs. M 0.8338451 0.759048

# ROC Curves

### Epi 패키지 이용

> library(Epi)
> ROC(form=cats$Sex~cats$Bwt, plot="ROC")

# ROC()는 Bwt, Hwt를 predictor로 넣는게 가능하네요(Sex ~ Bwt + Hwt)

> ROC(form=cats$Sex~cats$Bwt+cats$Hwt, plot="ROC")

### ROC 결과물

> data(cats, package="MASS")

> library(Epi)

> out = ROC(form=cats$Sex~cats$Bwt+cats$Hwt)

> out

$res

                        sens       spec       PV+       PV-    lr.eta

0.191243960703729 1.00000000 0.00000000 0.6736111       NaN        NA

0.193025369106466 0.97938144 0.00000000 0.6690141 0.0000000 0.1912440

0.194459564487476 0.97938144 0.02127660 0.6737589 0.3333333 0.1930254

0.202121627955293 0.97938144 0.04255319 0.6785714 0.5000000 0.1944596

0.255310074155313 0.97938144 0.06382979 0.6834532 0.6000000 0.2021216

0.255746729288057 0.97938144 0.08510638 0.6884058 0.6666667 0.2553101

0.257059632742871 0.97938144 0.10638298 0.6934307 0.7142857 0.2557467

.

.

.

0.994208761189084 0.04123711 1.00000000 1.0000000 0.3357143 0.9910133

0.994467173239425 0.03092784 1.00000000 1.0000000 0.3333333 0.9942088

0.995896529418839 0.02061856 1.00000000 1.0000000 0.3309859 0.9944672

0.99643076153607  0.01030928 1.00000000 1.0000000 0.3286713 0.9958965

                          0.00000000 1.00000000 1.0000000 0.3286713 0.9964308

$AUC

[1] 0.8329678

$lr

Call:  glm(formula = form, family = binomial)

Coefficients:

(Intercept)     cats$Bwt     cats$Hwt  

   -8.69896      3.55391      0.02295  

Degrees of Freedom: 143 Total (i.e. Null);  141 Residual

Null Deviance:      181.9 

Residual Deviance: 132.2        AIC: 138.2 

### logistic regression 결과와 동일

> out2=glm(factor(Sex)~Bwt+Hwt,data=cats,family=binomial)

### lr.eta는 p의 추정치

> fitted(out2)

### 추정치를 데이터 cats에 pred라는 변수로 추가. 이 pred가 lr.eta입니다. lr이 Logistic Regression의 약자인듯합니다.

> cats$pred=fitted(out2)

> head(cats)

  Sex Bwt Hwt      pred

1   F 2.0 7.0 0.1930254

2   F 2.0 7.4 0.1944596

3   F 2.0 9.5 0.2021216

4   F 2.1 7.2 0.2553101

5   F 2.1 7.3 0.2557467

6   F 2.1 7.6 0.2570596

[potassium]

아하. 그렇군요. 매번 감사합니다.

[김은주]

Epi 패키지 install한 후에 ROC 함수를 돌렸는데 (out1=ROC(form=na$group~na$sex+na$age+na$BMI+na$SBP+na$DBP+na$AST+na$ALT+na$Glucose+na$Uricacid+na$TG+na$HDL+na$LDL) ROC function을 찾을 수 없다는 에러가 뜨네요.. 뭐가 잘못된걸까요?

[안재형]

library(Epi) 하셨나요?

[kejjjj]

안재형 너무 기초적인 실수를햇네요.. R을 처음 다뤄봐서 ^^ 감사합니다!!

[안재형]

kejjjj 저두 가끔 빼먹어요^^