opImputation: A model-agnostic framework for dataset-specific selection of missing value imputation methods in pain-related numerical data

opImputation is an R package for comparing, benchmarking, and applying missing‑value imputation strategies to numerical tabular data. It is developed for biomedical and clinical research but is broadly applicable to any numerical dataset containing missing values.

Features

• Model‑agnostic benchmarking: Compare imputation methods from diverse algorithmic families
• Dataset‑specific selection: Automatically identify the best‑performing method for your data
• Automated imputation: Optionally produce a final imputed dataset using the top method
• Parallel processing: Efficient computation using the future framework (future.apply, progressr)
• Reproducible analysis: Seedable, standardized workflows
• Extensible integration: Add new methods or external benchmarking data easily

Installation

# From GitHub
if (!requireNamespace("devtools")) install.packages("devtools")
devtools::install_github("JornLotsch/opImputation")

# From CRAN (when available, uploaded to CRAN on October 24, 2025. Release pending incoming checks)
install.packages("opImputation")

Package metadata:

Usage

Basic example

library(opImputation)

# Load example data
data_iris <- iris[,1:4]

# Add some misisngs
set.seed(42)
for(i in 1:4) data_iris[sample(1:nrow(data_iris), 0.05*nrow(data_iris)), i] <- NA

# Basic comparison with a subset of methods
results <- compare_imputation_methods(
  data = data_iris,
  imputation_methods = c("mean", "median", "knn5", "rf_missForest"),
  n_iterations = 10,
  imputation_repetitions = 10,
  seed = 42
)

# Retrieve automatically generated final imputation
imputed_data <- results$imputed_data
print(results$method_used_for_imputation)

Main functions

compare_imputation_methods

Call

Returns

impute_missings

Call

Returns

A numeric data frame of the same dimensions and column names, with all missing values imputed.

Output and diagnostics

Performance evaluation is based on the standardized Δz (z‑delta) metric—
a robust measure of the absolute deviation between true and imputed values.
ABC (Activity‑Based Classification) categorizes imputation methods by their relative performance,
highlighting “A‑class” models as top performers.

Example output table from res_abc$df_abc_results[,1:3](generic dataset):

Legend:

• abc_score: zDelta values. Quantitative measure of imputation performance (higher = better).
• abc_category: ABC‑derived ranking class (“A” = top, “B” = medium, “C” = low).
• method: Name of the evaluated imputation algorithm.

“A‑class” methods (top seven in this example) represent the highest‑performing algorithms for the tested dataset.
Lower tiers correspond to progressively weaker or calibration‑only approaches.

Example summary plot

Diagnostic summary from the Iris dataset: ABC curves and variable‑specific Δz distributions (different dataset than that used in above table).

A: Standardized mean ranks for all imputation methods with ABC category coloring.
B: Mean standardized Δz deviations for diagnostic missings.
C: Variable‑level Δz distributions across methods.

When to use opImputation

• Biomedical or clinical datasets with incomplete numerical data
• Multivariate analysis or machine‑learning preprocessing
• Benchmarking and transparent method selection
• Fully automated, reproducible imputation pipelines

Citation

If you use opImputation, please cite:

Lötsch J, Ultsch A. (2025).
A model‑agnostic framework for dataset‑specific selection of missing value imputation methods in pain‑related numerical data.
Can J Pain (in minor revision)

Authors and license

• Jörn Lötsch (author, creator, maintainer)
• Alfred Ultsch (author)
• License: GPL‑3

About this project

opImputation is an R-based framework for automated, transparent, and reproducible selection of suitable missing‑value imputation methods for a given dataset. It provides a model‑agnostic approach that evaluates multiple candidate algorithms directly on the dataset to identify the most appropriate technique for that specific data structure, missingness pattern, and variable type.

The framework includes commonly used imputation families such as statistical, regression‑based, ensemble, tree‑based, and multiple‑imputation approaches. Its modular structure enables easy extension with additional or emerging methods, including deep‑learning‑based algorithms, ensuring long‑term flexibility across research domains.

Quantitative evaluation of candidate methods is based on computed ABC (cABC) analysis, which identifies statistically top‑performing algorithms through standardized performance metrics. For details on the theoretical background of cABC analysis, see: Ultsch A, Lötsch J. PLoS One. 2015; 10(6): e0129767.
Lötsch J, Ultsch A. Sci Rep. 2023; 13(1): 5470.