Using the missForest package

Hello there!

1 What is this all about?

In this short tutorial you will learn how to input NA data using a package called missForest. it uses random forests to input the values of your missing data based on data of other variables.

2 Load packages

First of all download and open the gapminder and missForest packages (and some other tools too)

knitr::opts_chunk$set(echo = TRUE)
library(missForest)

## Loading required package: randomForest

## randomForest 4.6-14

## Type rfNews() to see new features/changes/bug fixes.

## Loading required package: foreach

## Loading required package: itertools

## Loading required package: iterators

library(gapminder)
library(tidyverse)

## -- Attaching packages ------------------

## v ggplot2 3.3.0     v purrr   0.3.4
## v tibble  3.0.1     v dplyr   0.8.5
## v tidyr   1.0.3     v stringr 1.4.0
## v readr   1.3.1     v forcats 0.5.0

## -- Conflicts -- tidyverse_conflicts() --
## x purrr::accumulate() masks foreach::accumulate()
## x dplyr::combine()    masks randomForest::combine()
## x dplyr::filter()     masks stats::filter()
## x dplyr::lag()        masks stats::lag()
## x ggplot2::margin()   masks randomForest::margin()
## x purrr::when()       masks foreach::when()

library(skimr)

3 A simple example

Let’s start with a simple example, I have created three vectors, one which deliberately has one missing value.

variable_1=c(1:10)
variable_2=c(c(11:14), NA, c(16:20))
variable_3=c(21:30)
variable_1

##  [1]  1  2  3  4  5  6  7  8  9 10

variable_2

##  [1] 11 12 13 14 NA 16 17 18 19 20

variable_3

##  [1] 21 22 23 24 25 26 27 28 29 30

Let’s create a data-frame with those vectors, you can clearly see where the missing data is

dat_miss<- data.frame(variable_1,variable_2,variable_3)
dat_miss

By looking at the data we could clearly estimate that the missing value is variable_2=15. Let’s use the function missForest() to input the missing data.

dat_impo <- missForest(dat_miss)

##   missForest iteration 1 in progress...done!
##   missForest iteration 2 in progress...done!
##   missForest iteration 3 in progress...done!
##   missForest iteration 4 in progress...done!

class(dat_impo)

## [1] "missForest"

As you can see dat_impo becomes an object of the class "missForest". In order to get the new data frame we should do the following:

new_data <- dat_impo$ximp
new_data

As you can see the input value is variable_2=14.44 not too bad hey! we were expecting 15. Now let’s try to use this function for a more complex problem.

4 A not so simple Example

First let’s have a look at the gapminder data-set and let’s create a data set for all years for which we have GDP per capita data between 1980 and 2010… also let’s not consider population and life expectancy

head(gapminder)

gapm <- gapminder %>% 
  select(!c("continent","pop", "lifeExp")) %>% 
  filter(year %in% c(1980:2010)) %>% 
  spread(year, gdpPercap)
head(gapm)

Imagine that we have lost 10% of our values! let’s remove the name of the countries too

set.seed(110)
gapm<- gapm[,-1]
gap.mis <- prodNA(gapm, noNA = 0.1)
gap.mis <- as.data.frame(gap.mis)
head(gap.mis)

head(gapm)

Let’s double check the missing data, as you can see on average we have 17 GDP per capita missing values for each year.

skim(gap.mis)

Data summary

Name	gap.mis
Number of rows	142
Number of columns	6
_______________________
Column type frequency:
numeric	6
________________________
Group variables	None

Variable type: numeric

skim_variable	n_missing	complete_rate	mean	sd	p0	p25	p50	p75	p100	hist
1982	8	0.94	7361.07	7713.25	424.00	1313.30	4176.26	11652.52	33693.18	▇▂▂▁▁
1987	17	0.88	8008.55	8361.39	389.88	1361.94	4314.11	12037.27	31540.97	▇▂▁▁▁
1992	12	0.92	7943.65	8895.86	347.00	1219.89	4264.57	9604.72	34932.92	▇▁▁▂▁
1997	16	0.89	9223.21	10157.66	312.19	1366.84	5015.81	14073.69	41283.16	▇▂▁▁▁
2002	16	0.89	10386.03	11397.13	241.17	1647.27	5764.15	14542.65	44683.98	▇▁▁▂▁
2007	16	0.89	12115.81	13063.96	277.55	1624.84	7049.75	19166.11	49357.19	▇▂▁▂▁

Let’s impute missing values, using all the default parameters of the missForest function

set.seed(110)
gap.inp <- missForest(gap.mis, xtrue = gapm)

##   missForest iteration 1 in progress...done!
##   missForest iteration 2 in progress...done!
##   missForest iteration 3 in progress...done!
##   missForest iteration 4 in progress...done!

Again lets look at the new table

head(gap.inp$ximp)

How big are the errors of the imputation?

gap.inp$OOBerror

##     NRMSE 
## 0.1648565

gap.inp$error

##     NRMSE 
## 0.1660809

Not too bad, around 16%. NRME is The normalized root mean squared error, it is defined as:

\[\sqrt\frac{mean((Xtrue−Ximp)^2}{var(Xtrue))}\]

Check by yourself where the missing data is and what was it replaced with! moreover how similar it is compared with the original data!

head(gapm)

head(gap.mis)

head(gap.inp$ximp)

5 References

• https://www.rdocumentation.org/packages/missForest/versions/1.4/topics/missForest

• https://www.analyticsvidhya.com/blog/2016/03/tutorial-powerful-packages-imputing-missing-values/

• Stekhoven, D.J. and Buehlmann, P. (2012), ‘MissForest - nonparametric missing value imputation for mixed-type data’, Bioinformatics, 28(1) 2012, 112-118, doi: 10.1093/bioinformatics/btr597