ImputeBench

ImputeBench benchmarks single imputation methods for missing values through a standardized, empirical approach that enables context-specific selection and assessment of missing value imputation methods.

Key Features:

• Context-specific benchmarking: ImputeBench tailors benchmarking to reflect specific assumptions on missingness patterns, allowing application across diverse applied problems.
• Real data and simulation support: ImputeBench supports benchmarking via both real data and data simulation spanning latent factors, non-linearity, and multi-modality with interpretable, user-specifiable simulation parameters.
• Reproducible implementation: ImputeBench provides a reproducible R package implementation for missing value imputation selection and assessment.
• Method evaluation: ImputeBench enables selection of an imputation method for a given data set or benchmarking of a novel imputation method across a range of regimes.

Scientific Applications:

• Missing value imputation selection: ImputeBench selects optimal missing value imputation methods for specific data sets where the joint distribution of data and missingness patterns is unknown.
• Method comparison: ImputeBench compares existing imputation methods including k-nn, soft impute, missForest, and MICE in the context of real data from large-scale population-level studies.
• Novel method validation: ImputeBench benchmarks novel missing value imputation methods across diverse simulation regimes with controlled properties.

Methodology:

ImputeBench employs a standardized empirical benchmarking protocol with data simulation spanning latent factors, non-linearity, and multi-modality, implemented as a reproducible R package.