A General Multivariate Imputation Framework

Multivariate Expectation-Maximization (EM) based imputation framework that offers several different algorithms. These include regularisation methods like Lasso and Ridge regression, tree-based models and dimensionality reduction methods like PCA and PLS.

imputeR is an R package that provides a general framework for missing values imputation based on automated variable selection.

The main function impute inputs a matrix containing missing values and returns a complete data matrix using the variable selection functions provided as part of the package, or written by the user.

The package also offers many useful tools for imputation research based on impute. For example, the Detect function can be used to detect the variables' type in a given data matrix. guess can be used for naive imputation such as mean imputation, median imputation, majority imputation (for categorical variables only) and random imputation. SimIm function stands for "simulation for imputation". It accepts a complete matrix and randomly introduce some percentage of missing values into the matrix so imputation methods can be employed subsequently to impute this artificial missing data matrix. Because the true values are actually know so imputation accuracy can be easily calculated. This calls for the SimEval function that extends SimIm function, simulates a number of missing data matrices, applies a imputation method to these missing matrices and evaluate its performance. This enables the uncertainty of the imputation method to be obtained.

Reference

You can cite imputeR the following: General Multivariate Imputation Framework_. R package version 2.1, <URL: https://CRAN.R-project.org/package=imputeR>.

Version

2.1

License

GPL-3