Caret之Svm支持向量机

Posted 2021-04-26 OpeSource

本节主要介绍利用R进行Svm支持向量机分类，案例数据采用“german_credit”信息。

[cut] Read more

library(caret)  
library(pROC)  
library(ROCR)  

wd<-"G:/16支持向量/"
setwd(wd)
df<-read.csv("./german_credit_tra.csv",
             header=T,
             stringsAsFactors = FALSE)
german_credit<-as.data.frame(
             apply(df,2,as.numeric))
             
str(german_credit)
attach(german_credit)
german_credit$Creditability<-as.factor(
                             german_credit$Creditability)
levels(german_credit$Creditability) <- list(
                             no="0", yes="1") 

inTrain<-createDataPartition(
                   german_credit$Creditability,
                             p=0.70,list=FALSE)
                             
trainData<-german_credit[inTrain,]
testData<-german_credit[-inTrain,]
trainClass<-trainData$Creditability
testClass<-testData$Creditability
str(trainData)

set.seed(1)
control <- trainControl(
                    method = 'repeatedcv',
                    number = 10, 
                    repeats = 3,
                    classProbs = TRUE,
                    summaryFunction = twoClassSummary)
#####001svm radial model 高斯核
require(kernlab)
#宁可错杀不可放过，让本来信用不好的误判为信用好的例数减少
tuneRadial <- expand.grid(.sigma=c(0.1,0.2),
                      .C=c(0.1,0.3))
svm.model = train(Creditability ~ 
              AccountBalance + Duration +  
              PaymentStatusofPreviousCredit +Purpose +  
              ValueSavings, 
              
              data= trainData,
              method = "svmRadial",
              metric = "ROC",
              trControl = control,
              tuneGrid = tuneRadial)

summary(svm.model$finalModel)
svm.model$bestTune
svm.model$finalModel
svmImp<-varImp(svm.model,scale=FALSE)

svmImpprediction <- predict(svm.model,newdata=testData)
confusionMatrix(prediction, testClass)
              
#####002svm linear model 线性核
tuneLinear <- expand.grid(.C=c(0.3,0.5,0.7))
svm.model2 = train(Creditability ~ 
              AccountBalance + Duration + 
              PaymentStatusofPreviousCredit +Purpose + 
              ValueSavings, 
              
              data= trainData,
              method = "svmLinear",
              metric = "ROC",
              trControl = control,
              tuneGrid = tuneLinear)

summary(svm.model2$finalModel)
svm.model2$bestTune
svm.model2$finalModel
svmImp2<-varImp(svm.model2,scale=FALSE)
              
svmImp2prediction2 <- predict(svm.model2,newdata=testData)
confusionMatrix(prediction2, testClass)
detach(german_credit)
#混淆矩阵输出结果
            Reference
Prediction  no yes  
       no   38  25  
       yes  52 185
Accuracy : 0.7433