4.10 데이터마이닝

 

 

201301.ch2.r.mining4_10수업.eng.hwp

 

 

4_10마이닝.R

 

 

 

?read.table

 

dataset<-read.csv(file="D:/abc/dm.hw.01.dataset.01.csv")

 

?glm

 

 

Y<-dataset[,1]
X<-as.matrix(dataset[,-1])

 

length(Y)

dim(X)

 

 

plot(X[,1],Y)
plot(X[,10],Y)     # 선형모양이 아닌것처럼 보여도 데이터에서 제외하면 안됨

 

 

Y<-(Y>0)*1

 

myfit<-summary(glm(Y~X,family="binomial")) #?glm , ?family 확인하기

myfit

 

summary(myfit)

 

glm(formula = Y ~ X, family = "binomial")

Deviance Residuals:
     Min        1Q    Median        3Q       Max 
-1.77621  -0.09955   0.00015   0.10863   1.80266 

Coefficients:
            Estimate Std. Error z value Pr(>|z|)        #pr(>z) ---> p밸류
(Intercept)   0.5261     0.4483   1.173 0.240618   
XX1           5.1485     1.5909   3.236 0.001211 **
XX2           3.6369     1.0841   3.355 0.000795 ***
XX3           4.9238     1.5072   3.267 0.001087 **
XX4           2.3151     0.9106   2.542 0.011009 * 
XX5           2.5618     0.9186   2.789 0.005290 **
XX6          -0.7336     0.5726  -1.281 0.200161   
XX7           1.6666     0.7810   2.134 0.032849 * 
XX8           1.3549     0.6470   2.094 0.036238 * 
XX9          -0.5826     0.5456  -1.068 0.285604   
XX10          0.3062     0.4082   0.750 0.453197   
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

(Dispersion parameter for binomial family taken to be 1)

    Null deviance: 138.589  on 99  degrees of freedom
Residual deviance:  39.849  on 89  degrees of freedom
AIC: 61.849

Number of Fisher Scoring iterations: 8

 

 

 

names(myfit)  # 계수 확인?

 

myfit$coef
myfit$coef[,1]
X[1,]

 

 

B<-myfit$coef[,1]
X0<-c(1,X[1,])

L0<-t(X0)%*%B

exp(L0)/(1+exp(L0))

L1<-cbind(1,X)%*%B

hat.y<-(exp(L1)/(1+exp(L1))>0.5)*1

 

e<-cbind(Y,hat.y)   # Y와 hat.y 비교
e

 

sen<-sum(hat.y[Y==1])/sum(Y)   #1인 것의 비율
spc<-sum(hat.y[Y==0])/(length(Y)-sum(Y))