R 语言中的一种 Class 分类。 生成混淆矩阵时我做错了什么？

Question

I am trying to understand and implement classifiers A class in R is based on several UCIs and one of them ( http://archive.ics.uci.edu/ml/datasets/Chronic_Kidney_Disease ).

When trying to print a confusion matrix you are giving the error “all arguments must have the same length”.

What am I doing wrong?

library(caret)
library(dplyr)
library(e1071)
library(NLP)
library(tm)

ds = read.csv('kidney_disease.csv', 
              header = TRUE)

#Remover colunas inutiliz?veis              
ds <- subset(ds, select = -c(age), classification =='ckd' )

x <- subset(ds, select = -classification) #make x variables
y <- ds$classification #make y variable(dependent)

# test on the whole set
#pred <- predict(model, subset(ds, select=-classification))


trainPositive<-x
testnegative<-y

inTrain<-createDataPartition(1:nrow(trainPositive),p=0.6,list=FALSE)

trainpredictors<-trainPositive[inTrain,1:4]
trainLabels<-trainPositive[inTrain,6]

testPositive<-trainPositive[-inTrain,]
testPosNeg<-rbind(testPositive,testnegative)

testpredictors<-testPosNeg[,1:4]
testLabels<-testPosNeg[,6]

svm.model<-svm(trainpredictors,y=NULL,
               type='one-classification',
               nu=0.10,
               scale=TRUE,
               kernel="radial")

svm.predtrain<-predict(svm.model,trainpredictors)
svm.predtest<-predict(svm.model,testpredictors)

# confusionMatrixTable<-table(Predicted=svm.pred,Reference=testLabels)
# confusionMatrix(confusionMatrixTable,positive='TRUE')

confTrain <- table(Predicted=svm.predtrain,Reference=trainLabels)
confTest <- table(Predicted=svm.predtest,Reference=testLabels)

confusionMatrix(confTest,positive='TRUE')


print(confTrain)
print(confTest)

#grid

Here are some of the first lines of the dataset I'm using:

 id bp    sg al su    rbc       pc        pcc         ba bgr bu  sc sod pot hemo pcv   wc
1  0 80 1.020  1  0          normal notpresent notpresent 121 36 1.2  NA  NA 15.4  44 7800
2  1 50 1.020  4  0          normal notpresent notpresent  NA 18 0.8  NA  NA 11.3  38 6000
3  2 80 1.010  2  3 normal   normal notpresent notpresent 423 53 1.8  NA  NA  9.6  31 7500
4  3 70 1.005  4  0 normal abnormal    present notpresent 117 56 3.8 111 2.5 11.2  32 6700
5  4 80 1.010  2  0 normal   normal notpresent notpresent 106 26 1.4  NA  NA 11.6  35 7300
6  5 90 1.015  3  0                 notpresent notpresent  74 25 1.1 142 3.2 12.2  39 7800
   rc htn  dm cad appet  pe ane classification
1 5.2 yes yes  no  good  no  no            ckd
2      no  no  no  good  no  no            ckd
3      no yes  no  poor  no yes            ckd
4 3.9 yes  no  no  poor yes yes            ckd
5 4.6  no  no  no  good  no  no            ckd
6 4.4 yes yes  no  good yes  no            ckd

The error log:

> confTrain <- table (Predicted = svm.predtrain, Reference = trainLabels)
Table error (Predicted = svm.predtrain, Reference = trainLabels):
all arguments must be the same length
> confTest <- table (Predicted = svm.predtest, Reference = testLabels)
Table error (expected = svm.predtest, reference = testLabels):
all arguments must be the same length
>
> confusionMatrix (confTest, positive = 'TRUE')
ConfusionMatrix error (confTest, positive = "TRUE"):
'confTest' object not found
>
>
> print (confTrain)
Printing error (confTrain): object 'confTrain' not found
> print (confTest)
Printing error (confTest): object 'confTest' not found

Answer 1

I see a number of issues. First it seems that a lot of your data is of class character rather than numeric, which is required by the classifier. Let's pick some columns and convert to numeric. I will use data.table because fread is very convenient.

library(caret)
library(e1071)
library(data.table)
setDT(ds)
#Choose columns
mycols <- c("id","bp","sg","al","su")
#Convert to numeric
ds[,(mycols) := lapply(.SD, as.numeric),.SDcols = mycols]

#Convert classification to logical
data <- ds[,.(bp,sg,al,su,classification = ds$classification == "ckd")]
data
     bp    sg al su classification
  1: 80 1.020  1  0           TRUE
  2: 50 1.020  4  0           TRUE
  3: 80 1.010  2  3           TRUE
  4: 70 1.005  4  0           TRUE
  5: 80 1.010  2  0           TRUE
 ---                              
396: 80 1.020  0  0          FALSE
397: 70 1.025  0  0          FALSE
398: 80 1.020  0  0          FALSE
399: 60 1.025  0  0          FALSE
400: 80 1.025  0  0          FALSE

Once the data is cleaned up, you can sample a training and test set with createDataPartition as in your original code.

#Sample data for training and test set
inTrain<-createDataPartition(1:nrow(data),p=0.6,list=FALSE)
train<- data[inTrain,]
test <- data[-inTrain,]

Then we can create the model and make the predictions.

svm.model<-svm(classification ~ bp + sg + al + su, data = train,
               type='one-classification',
               nu=0.10,
               scale=TRUE,
               kernel="radial")

#Perform predictions 
svm.predtrain<-predict(svm.model,train)
svm.predtest<-predict(svm.model,test)

Your main issue with the cross table was that the model can only predict for cases that don't have any NA s, so you have to subset the classification levels to those with predictions. Then you can evaluate confusionMatrix :

confTrain <- table(Predicted=svm.predtrain,
                   Reference=train$classification[as.integer(names(svm.predtrain))])
confTest <- table(Predicted=svm.predtest,
                  Reference=test$classification[as.integer(names(svm.predtest))])

confusionMatrix(confTest,positive='TRUE')

Confusion Matrix and Statistics

         Reference
Predicted FALSE TRUE
    FALSE     0   17
    TRUE     55   64

               Accuracy : 0.4706         
                 95% CI : (0.3845, 0.558)
    No Information Rate : 0.5956         
    P-Value [Acc > NIR] : 0.9988         

                  Kappa : -0.2361        

 Mcnemar's Test P-Value : 1.298e-05      

            Sensitivity : 0.7901         
            Specificity : 0.0000         
         Pos Pred Value : 0.5378         
         Neg Pred Value : 0.0000         
             Prevalence : 0.5956         
         Detection Rate : 0.4706         
   Detection Prevalence : 0.8750         
      Balanced Accuracy : 0.3951         

       'Positive' Class : TRUE           

Data

library(archive)
library(data.table)
tf1 <- tempfile(fileext = ".rar")
#Download data file
download.file("http://archive.ics.uci.edu/ml/machine-learning-databases/00336/Chronic_Kidney_Disease.rar", tf1)
tf2 <- tempfile()
#Un-rar file
archive_extract(tf1, tf2)
#Read in data
ds <- fread(paste0(tf2,"/Chronic_Kidney_Disease/chronic_kidney_disease.arff"), fill = TRUE, skip = "48")
#Remove erroneous last column
ds[,V26:= NULL]
#Set column names (from header)
setnames(ds,c("id","bp","sg","al","su","rbc","pc","pcc","ba","bgr","bu","sc","sod","pot","hemo","pcv","wc","rc","htn","dm","cad","appet","pe","ane","classification"))
#Replace "?" with NA
ds[ds == "?"] <- NA