r - Prediction for new observation in knn - r

Question

I am trying to make an application which would predict prices based on users input. How can I predict the response for new values?
I have tried to do the following:
1. Add a new observation to the dataset
2. Train knn on all of the observations but the new one
3. Test knn on the new observation
But the prediction changes when I put different values of the response variable into the new observation so it doesn't seem to work.
Let's say the data has 100 observations of 7 variables.
This would be the code I have tried.
data <- rbind(data, c(1,2,3,4,5,6,7))
prediction <- knn.reg(data[1:100,], test = dataset[101,],
data[1:100,]$response_variable, k = 8, algorithm="kd_tree")
prediction$pred
Thank you in advance for your help.

Answer 1

For one thing, you have not defined dataset. I am guessing your code is meant to read:
dataset <- rbind(data, c(1,2,3,4,5,6,7))
prediction <- knn.reg(dataset[1:100,], test = dataset[101,],
y = dataset[1:100,]$response_variable, k = 8, algorithm="kd_tree")
prediction$pred
In any case, it seems that you are not supposed to include the response variable as a column in your training and test sets (I found this out by playing around with the knn.reg function.) So, if your response variable was the 7th column of data then you could do this instead
dataset <- rbind(data, c(1,2,3,4,5,6,7))
prediction <- knn.reg(dataset[1:100,-7], test = dataset[101,-7],
y = dataset[1:100,]$response_variable, k = 8, algorithm="kd_tree")
prediction$pred
For example, here is a test case with some made-up data.
set.seed(1)
data <- data.frame(matrix(sample(1:7, 700, replace=T), nr=100))
colnames(data)[7] <- "response_variable"
dataset <- rbind(data, c(1,2,3,4,5,6,7))
prediction <- knn.reg(dataset[1:100,-7], test = dataset[101,-7],
dataset[1:100,]$response_variable, k = 8, algorithm="kd_tree")
prediction$pred