How to Handle Missing Data in R with simputation

Written by Matt Dancho

This article is part of a R-Tips Weekly, a weekly video tutorial that shows you step-by-step how to do common R coding tasks.

Missing values used to drive me nuts… until I learned how to impute them! In 10-minutes, learn how to visualize and impute in R using ggplot dplyr and 3 more packages to simple imputation.

Here are the links to get set up. 👇

Handling missing values

We’re going to kick the tires on 3 key packages:

visdat - For quickly visualizing data
naniar - For working with NA’s (missing data)
simputation - For simple imputation (converting missing data to values)

So let’s get started!

Visualizing Missing Data
Using vis_miss(), gg_miss_upset() and geom_miss_point()

Quickly Skim Missing Data

It doesn’t get any easier than this. Simply use visdat::vis_miss() to visualize the missing data. We can see Ozone and Solar.R are the offenders.

Identify Interactions in Column Missingness

Use Case: It often makes sense to evaluate the interactions between columns containing missing data. We can use an “upset” plot for this.

Start with a good question:

“Is it often that we have both Ozone and Solar.R missing at the same time?”

We can answer this with gg_miss_upset(). We can see that 2 of 5 Solar.R (40%) happen at the same observation that Ozone is missing. Might want to check for IOT sensor issues!

Visualize Missing Observations in a Scatter Plot

Use Case: This is a great before/after visual.

For our final exploratory plot, let’s plot the missing data using geom_miss_point(). It works just like geom_point(), but plots where the missing data are located in addition to the non-missing data.

Before Imputation. Using geom_miss_point()

Imputation
impute_rf()

The simputation library comes with a host of impute*()_ functions. We’ll focus on impute_rf(), which implements a random forest to do the imputation.

This imputes the NA’s, replacing the missing Ozone and Solar.R data. We can see the missing data follows the distribution of the non-missing data in the updated scatter plot.