Predictive Hacks

# How to Impute Missing Values in R

In the real data world, it is quite common to deal with Missing Values (known as NAs). Sometimes, there is a need to impute the missing values where the most common approaches are:

• Numerical Data: Impute Missing Values with mean or median
• Categorical Data: Impute Missing Values with mode

Let’s give an example of how we can impute dynamically depending on the data type.

```library(tidyverse)

df<-tibble(id=seq(1,10), ColumnA=c(10,9,8,7,NA,NA,20,15,12,NA),
ColumnB=factor(c("A","B","A","A","","B","A","B","","A")),
ColumnC=factor(c("","BB","CC","BB","BB","CC","AA","BB","","AA")),
ColumnD=c(NA,20,18,22,18,17,19,NA,17,23)
)

df

```
```# A tibble: 10 x 5
id ColumnA ColumnB ColumnC ColumnD
<int>   <dbl> <fct>   <fct>     <dbl>
1     1      10 "A"     ""           NA
2     2       9 "B"     "BB"         20
3     3       8 "A"     "CC"         18
4     4       7 "A"     "BB"         22
5     5      NA ""      "BB"         18
6     6      NA "B"     "CC"         17
7     7      20 "A"     "AA"         19
8     8      15 "B"     "BB"         NA
9     9      12 ""      ""           17
10    10      NA "A"     "AA"         23
```

For the Categorical Variables, we are going to apply the “mode” function which we have to build it since it is not provided by R.

```getmode <- function(v){
v=v[nchar(as.character(v))>0]
uniqv <- unique(v)
uniqv[which.max(tabulate(match(v, uniqv)))]
}

```

Now that we have the “mode” function we are ready to impute the missing values of a dataframe depending on the data type of the columns. Thus, if the column data type is “numeric” we will impute it with the “mean” otherwise with the “mode“. Notice that in our script we take into account the column names and “dplyr” package requires a special notation (!!cols : = !!rlang::sym(colname)) of selecting dynamically the column names.

```for (cols in colnames(df)) {
if (cols %in% names(df[,sapply(df, is.numeric)])) {
df<-df%>%mutate(!!cols := replace(!!rlang::sym(cols), is.na(!!rlang::sym(cols)), mean(!!rlang::sym(cols), na.rm=TRUE)))

}
else {

df<-df%>%mutate(!!cols := replace(!!rlang::sym(cols), !!rlang::sym(cols)=="", getmode(!!rlang::sym(cols))))

}
}

df

```
```> df
# A tibble: 10 x 5
id ColumnA ColumnB ColumnC ColumnD
<dbl>   <dbl> <fct>   <fct>     <dbl>
1     1    10   A       BB         19.2
2     2     9   B       BB         20
3     3     8   A       CC         18
4     4     7   A       BB         22
5     5    11.6 A       BB         18
6     6    11.6 B       CC         17
7     7    20   A       AA         19
8     8    15   B       BB         19.2
9     9    12   A       BB         17
10    10    11.6 A       AA         23
```

Voilà! The missing values have been imputed!

### 3 thoughts on “How to Impute Missing Values in R”

1. I would encourage you to look into multiple imputation (the mice package is really good). Mean and mode imputation is almost worst than doing nothing as it will increase the bias and shrink error estimates.

• Thank you Jason. I agree that is risky to impute missing values with mean and median when it comes to predictive models. Our goal here is to show how you can do some things. Regarding the `mice` package. It returns the missing values based on probabilistic models which also can be biased.

2. Hello.
This gives no error but no imputation either. Can anyone help?
E. Borsting

>
> library(tidyverse)
>
>
>
> class(df)
[1] “data.frame”
> df
5b c…2 a…3 a…4 b…5 c…6 b…7 c…8 b…9 a…10
1 64 c b c b a c c b c
2 ab4 c b b b c c a
3 3a c b b b b a b a b
4 5c c c b c b b b b c
5 a74 c b b b b a c a c
6 a99 c a c a b b c b b
7 aa8 c a a a c
8 a68 c a b b b c c a c
9 b4b c c b b b a c a c
10 b43 c a a b b b b a a
>
> getmode 0]
+ uniqv
>
> for (cols in colnames(df)) {
+ if (cols %in% names(df[,sapply(df, is.numeric)])) {
+ df%mutate(!!cols := replace(!!rlang::sym(cols), is.na(!!rlang::sym(cols)), mean(!!rlang::sym(cols), na.rm=TRUE)))
+
+ }
+ else {
+
+ df%mutate(!!cols := replace(!!rlang::sym(cols), !!rlang::sym(cols)==””, getmode(!!rlang::sym(cols))))
+
+ }
+ }
>
> df
5b c…2 a…3 a…4 b…5 c…6 b…7 c…8 b…9 a…10
1 64 c b c b a c c b c
2 ab4 c b b b c c a
3 3a c b b b b a b a b
4 5c c c b c b b b b c
5 a74 c b b b b a c a c
6 a99 c a c a b b c b b
7 aa8 c a a a c
8 a68 c a b b b c c a c
9 b4b c c b b b a c a c
10 b43 c a a b b b b a a

### Get updates and learn from the best

Python

#### Creating Dynamic Forms with Streamlit: A Step-by-Step Guide

In this blog post, we’ll teach you how to create dynamic forms based on user input using Streamlit’s session state

Python

#### How to Connect Wikipedia with ChatGPT and LangChain

ChatGPT’s knowledge is limited to its training data, which has the cutoff year of 2021. This implies that we cannot