README.md

---

Overview of missCompare

missCompare is a missing data imputation pipeline that will guide you through your missing data problem. A range of functions will help you select what could be the most ideal algorithm for your data and provide an easy way to impute missing datapoints in your dataset.

The missCompare pipeline

You will find a detailed manual in the missCompare vignette:

install.packages("missCompare")
library(missCompare)
vignette("misscompare")

1. Cleaning your data using missCompare::clean()
2. Extracting information on dimensions, missingness, correlations and variables, plotting missing data using missCompare::get_data()
3. Imputation - simulated data:

• simulating full data with no missingness using metadata from the previous step (resembling your original data) using missCompare::simulated()
• spiking in missing data in distinct missing data patterns using missCompare::all_patterns(). These patterns are:
  • missing completely at random (MCAR) - missCompare::MCAR() - missing data occurrence random
  • missing at random (MAR) - missCompare::MAR() - missing data occurrence correlates with other variables' values (univariate solution in missCompare)
  • missing not at random (MNAR) - missCompare::MNAR() - missing data occurrence correlates with variables' own values
  • missing in assumed pattern (MAP) - missCompare::MAP() - a combination of the previous three, where the user can define a pattern per variable
• imputing missing data, obtaining imputation metrics (root mean squared errors - RMSE, mean absolute error - MAE, Kolmogorov-Smirnov test statistic D for equal distributions, computation time) per method and plotting results using missCompare::impute_simulated()

4. Imputing your data - After the previous step, you will have a general idea about what are the best performing algorithms for your data structure (size, degree of correlation between variables). In this step, you can impute your original data with your chosen algorithm(s) using missCompare::impute_data()
5. Post imputation diagnostics will give an informative assessment on how the imputation changed your data structure (e.g. variable means, distributions, clusters, correlations). The function here is missCompare::post_imp_diag()

Installation

You can install the released version of missCompare from CRAN with:

install.packages("missCompare")

Usage

Loading library and sandbox data

library(missCompare)
data("clindata_miss")

Cleaning

cleaned <- missCompare::clean(clindata_miss,
                              var_removal_threshold = 0.5, 
                              ind_removal_threshold = 1,
                              missingness_coding = -9)

Extracting data

metadata <- missCompare::get_data(cleaned,
                                  matrixplot_sort = T,
                                  plot_transform = T)

Imputation - simulation framework

missCompare::impute_simulated(rownum = metadata$Rows,
                              colnum = metadata$Columns, 
                              cormat = metadata$Corr_matrix,
                              MD_pattern = metadata$MD_Pattern,
                              NA_fraction = metadata$Fraction_missingness,
                              min_PDM = 10,
                              n.iter = 50, 
                              assumed_pattern = NA)

Computation time comparison

RMSE comparison

KS comparison

Imputation of data

imputed <- missCompare::impute_data(cleaned, 
                         scale = T, 
                         n.iter = 10, 
                         sel_method = c(1:16))

Post imputation diagnostics

diag <- missCompare::post_imp_diag(cleaned,
                                   imputed$mean_imputation[[1]],
                                   scale=T, 
                                   n.boot = 100)

Post imputation diagnostics - distributions of original and imputed values for a random variable

Post imputation diagnostics - variable clusters in the original and imputed datasets

Post imputation diagnostics - comparison of variable-pair correlations

Issues, questions

In case you need help or advice on your missing data problem or you need help with the missCompare package, please e-mail the authors. If you would like to report an issue, please do so in a reproducible example at the missCompare GitHub page.