KNN model gets 100% accuracy - should I trust the results?

Asked Sep 11 '18 at 14:39

Active Sep 11 '18 at 15:07

Viewed 1,114 times

So I'm playing with the iris dataset and no matter what my value of k in the knn algorithm, I'm getting 100% accuracy. Have I gone wrong somewhere? Here's my code using the in-built iris data frame.

library(caret)
set.seed(52)
irissplit <- createDataPartition(iris$Species, p=.8, list=FALSE, times=1)
train <- iris[irissplit, ]
test <- iris[-irissplit, ]

library(class)
model_knn <- list()
accuracy_knn <- numeric()
for (i in 1:10) {
 model_knn[[i]] <- knn(train[,-5], test[,-5], train$Species, k=i, prob=TRUE) 
 accuracy_knn[i] <- sum(model_knn[[i]]==test$Species)/length(test$Species)*100
}
accuracy_knn

edited Sep 11 '18 at 14:44

G5W

36,531
10
47
80

asked Sep 11 '18 at 14:39

stixmcvix

1

Where does `createDataPartition` come from? – Axeman Sep 11 '18 at 14:40
Its from the caret package. It partitions the dataset ready for splitting out into training and test sets. – stixmcvix Sep 11 '18 at 15:06
No, never trust a 100% accuracy model. In fact, 100% accuracy always means that you've made a mistake in some step – antonioACR1 Sep 11 '18 at 15:10
2

By changing the `set.seed` it gives something different. The code seems correct. – RLave Sep 11 '18 at 15:17
Oh wow! I somehow managed to pick the optimum seed! Thanks for the comment though, glad to see I'm not losing my marbles. – stixmcvix Sep 11 '18 at 15:25

KNN model gets 100% accuracy - should I trust the results?

0 Answers0