I am using the class package in order to use the KNN algorithm. I am also using the ROCR package to calculate the AUC value.

knn_one<-knn(train, test, train$Digit, k=1)

To calculate the AUC value for another method, e.g. classification trees, I used these series of commands:

treeTrain_Pred<-predict(Tree_Train, test , type = "prob")[,2]
Pred<-prediction(treeTrain_Pred, test$Digit)
Perf<-performance(Pred, "auc")
[[1]]

However, when I try

knn_one = predict(knn_one, test, type="prob")[,2]

I get the following error:

Error in UseMethod("predict") :
no applicable method for 'predict' applied to an object of class "factor"

How can I fix this and obtain an AUC value for my KNN function?

3

1 Answer

There is no predict method for knn models, instead you train and receive predictions as part of a single call. Example on sonar data:

library(mlbench)
data(Sonar) 

create data partition:

set.seed(1)
tr_ind <- sample(1:nrow(Sonar), 150)
train <- Sonar[tr_ind,]
test <- Sonar[-tr_ind,]
mod <- class::knn(cl = train$Class, test = test[,1:60], train = train[,1:60], k = 5, prob = TRUE)

Now the probability of the predictions are in:

attributes(mod)$prob
library(pROC)
roc(test$Class, attributes(mod)$prob)
#output
Call:
roc.default(response = test$Class, predictor = attributes(mod)$prob)
Data: attributes(mod)$prob in 30 controls (test$Class M) < 28 cases (test$Class R).
Area under the curve: 0.4667
plot(roc(test$Class, attributes(mod)$prob), print.thres = T, print.auc=T)

lets try with k = 4

mod <- class::knn(cl = train$Class, test = test[,1:60], train = train[,1:60], k = 4, prob = TRUE)
plot(roc(test$Class, attributes(mod)$prob), print.thres = T, print.auc = T, print.auc.y = 0.2)

4

Your Answer

Sign up or log in

Sign up using Google Sign up using Facebook Sign up using Email and Password Submit

Post as a guest

Post Your Answer Discard

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy