This chapter describes how the Microimpute package allows you to compare different imputation methods using preprocessing, cross-validation, metric comparison, and evaluation tools.
The benchmarking functionality enables systematically comparing multiple imputation models using a common dataset, allowing for robust evaluation of their performance. It supports cross-validation to diagnose overfitting and measure performance on training data leveraging the availability of ground truth. By assessing accuracy of numeric imputation across various quantiles, it is possible to gain a more comprehensive understanding of how each method performs across different levels of the distribution. Categorical imputation is assessed with log loss. This process is further supported by visualizations that highlight differences between approaches, making it easy to identify which imputation methods perform best under specific conditions. Predictor evaluation tools are also available to inform decision-making when setting up the imputation task.