R 多元线性回归

install.packages("Hmisc")
install.packages("mice")
library(Hmisc)
library(mice)
data_1<-impute(ma317projectdata_2$X2012SP.DYN.LE00.IN,mean)
#对结果进行填充
ma317projectdata_2$X2012SP.DYN.LE00.IN[is.na(ma317projectdata_2$X2012SP.DYN.LE00.IN)]<-mean(ma317projectdata_2$X2012SP.DYN.LE00.IN, na.rm = T)


mydata<-as.data.frame(ma317projectdata_2)
#相关性检验
cov(mydata)
# X2012SH.XPD.PCAP X2012SL.UEM.1524.ZS X2012GC.BAL.CASH.GD.ZS X2012SP.DYN.LE00.IN
# X2012SH.XPD.PCAP           3.006310e+06           40.538378             -9.2014918        7767.9251172
# X2012SL.UEM.1524.ZS        4.053838e+01          111.615457             -6.6161710          17.1944126
# X2012GC.BAL.CASH.GD.ZS    -9.201492e+00           -6.616171             16.1245419          -0.6002212
# X2012SP.DYN.LE00.IN        7.767925e+03           17.194413             -0.6002212          74.9118626


#建立模型
mydata_Linear_regression<-lm(mydata$X2012SP.DYN.LE00.IN~.,data=mydata)

#模型评估
mydata_Linear_regression


# Call:
#   lm(formula = mydata$X2012SP.DYN.LE00.IN ~ ., data = mydata)
# 
# Coefficients:
#   (Intercept)        X2012SH.XPD.PCAP     X2012SL.UEM.1524.ZS  X2012GC.BAL.CASH.GD.ZS  
# 64.914619                0.002582                0.154758                0.027749  


summary(mydata_Linear_regression)
# Call:
#   lm(formula = mydata$X2012SP.DYN.LE00.IN ~ ., data = mydata)
# 
# Residuals:
#   Min      1Q  Median      3Q     Max 
# -24.074  -3.137   1.337   4.715  14.328 

# Coefficients:
#   Estimate Std. Error t value Pr(>|t|)    
# (Intercept)            6.491e+01  9.806e-01  66.198  < 2e-16 ***
#   X2012SH.XPD.PCAP       2.582e-03  2.662e-04   9.697  < 2e-16 ***
#   X2012SL.UEM.1524.ZS    1.548e-01  4.424e-02   3.498 0.000556 ***
#   X2012GC.BAL.CASH.GD.ZS 2.775e-02  1.164e-01   0.238 0.811755    
# ---
#   Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1
# 
# Residual standard error: 7.27 on 245 degrees of freedom
# Multiple R-squared:  0.303,	Adjusted R-squared:  0.2945 
# F-statistic: 35.51 on 3 and 245 DF,  p-value: < 2.2e-16


library(car)
scatterplotMatrix(mydata,spread=FALSE)
#图一 各变量线性相关图


# 缺点：虽然它可能会找到一个好的模型，但是不能保证模型就是最佳型，因为不是每一个可能的模型都被评价了。
# 全子集回归
# 即所有可能的模型都会被检验。
# 全子集回归可用leaps包中的regsubsets()函数实现。
#具体的regsubsets 的图可以表示出什么意思你可以自己查查相关的解释

install.packages('leaps')
library(leaps)
leaps<-regsubsets(mydata$X2012SP.DYN.LE00.IN~.,data=mydata,nbest = 3)
plot(leaps,scale = 'adjr2')
#图二 regsubsets图


coef(mydata_Linear_regression)

# coef(mydata_Linear_regression)
# (Intercept)       X2012SH.XPD.PCAP    X2012SL.UEM.1524.ZS X2012GC.BAL.CASH.GD.ZS 
# 64.914619203            0.002581872            0.154757591            0.027748915 
#通过这几种方法，我们都可以明显的看出预期寿命与 X2012SL.UEM.1524.ZS相关性较大，与其它因素相关性较小。

#回归诊断
confint(mydata_Linear_regression)


# confint(mydata_Linear_regression)
# 2.5 %       97.5 %
#   (Intercept)            62.983122459 66.846115947
# X2012SH.XPD.PCAP        0.002057446  0.003106298
# X2012SL.UEM.1524.ZS     0.067624047  0.241891135
# X2012GC.BAL.CASH.GD.ZS -0.201497827  0.256995658

#标记异常值

qqPlot(mydata_Linear_regression,labels = row.names(mydata),id.method = 'identify',simulate = T)
#图三 有些数据就可以抛弃不用


#预测模型 test  是你的预测数据
p<-predict(mydata_Linear_regression,test)

write.csv(mydata,file="C:/Users/wwq/Documents/data_pre.csv")
getwd()
猜你喜欢
