kNN tricks & tips II: sorting variables
As the k-Nearest Neighbors algorithm loops over the variables in the data to impute them, it computes distances between observations using other variables, some of which have already been imputed in the previous steps. This means that if the variables located earlier in the data have a lot of missing values, then the subsequent distance calculation is based on a lot of imputed values. This introduces noise to the distance calculation.
For this reason, it is a good practice to sort the variables increasingly by the number of missing values before performing kNN imputation. This way, each distance calculation is based on as much observed data and as little imputed data as possible.
Let's try this out on the tao data!
This exercise is part of the course
Handling Missing Data with Imputations in R
Exercise instructions
- Calculate the number of missing values in each column of
taoin the first part of the pipeline. - Then, sort the variables increasingly according to the number of missing values, extract their names and assign the result to
vars_by_NAs. - Use
select()to reordertaovariables using the order saved invars_by_NAs. - Perform k-Nearest Neighbors imputation on the reordered data and assign the result to
tao_imp.
Hands-on interactive exercise
Have a go at this exercise by completing this sample code.
# Get tao variable names sorted by number of NAs
vars_by_NAs <- tao %>%
___ %>%
colSums() %>%
sort(decreasing = ___) %>%
names()
# Sort tao variables and feed it to kNN imputation
tao_imp <- tao %>%
select(___) %>%
___()
tao_imp %>%
select(sea_surface_temp, humidity, humidity_imp) %>%
marginplot(delimiter = "imp")