Skip to contents

Hybrid Imputation: Random Forest + LCMD

hybrid_imputation.Rd

Performs hybrid imputation on selected target columns of a data frame by combining Random Forest (RF) imputation via missForest and left-censored missing data (LCMD) imputation via imputeLCMD. Only non-QC rows are imputed. QC rows are excluded from model fitting, OOB error estimation, and imputation, and are returned unchanged. Non-target columns are also returned unchanged.

Usage

hybrid_imputation(
  df,
  target_cols = NULL,
  is_qc = NULL,
  method = c("RF-LCMD"),
  oobe_threshold = 0.1,
  control_RF = list(),
  control_LCMD = list()
)

Arguments

df

A data frame with samples (rows) and features (columns).

target_cols

Character vector of column names, or a single regular expression, identifying columns to impute. Only these columns are passed to the RF and LCMD imputation routines and only these columns can be modified in the returned data frames. Non-target columns are retained but are not imputed or used for method selection. If NULL, target columns are resolved automatically using resolve_target_cols().

is_qc

A logical vector indicating which rows are QC samples. Must match nrow(df).

method

Imputation strategy to use (currently only "RF-LCMD" supported).

oobe_threshold

Numeric. Features with OOBE below this threshold will use RF, others will use LCMD.

control_RF

A named list of control arguments for missForest::missForest(). Also supports n_cores (internal).

control_LCMD

A named list of control arguments for imputeLCMD::impute.MAR.MNAR(), including mode = "overall" or "column-wise".

Value

A named list with the following components:

hybrid_rf_lcmd

The fully imputed data frame combining RF and LCMD decisions.

rf

The RF-imputed data frame (non-QC rows only, in full column structure).

lcmd

The LCMD-imputed data frame (non-QC rows only, in full column structure).

oob

A named numeric vector of feature-level OOB errors from RF.

Details

For each target column, the final imputed values are selected according to the feature-level out-of-bag error (OOBE) from the RF model. Target columns with RF OOBE strictly below oobe_threshold use RF-imputed values, whereas target columns with RF OOBE greater than or equal to oobe_threshold use LCMD-imputed values.

In addition to the hybrid result, the function also returns the complete RF and LCMD imputed data frames. This allows users to inspect feature-level method choices, perform sensitivity analyses, compare alternative imputation strategies, or generate new hybrid imputed data frames using different OOBE thresholds without rerunning the imputation procedures.

See also

run_rf_imputation, run_lcmd_imputation