Missing values Imputation with five fold cross validation using python

Question

I have a dataset of 165 instances and 49 features with target 1 and 0. This dataset has missing values so i am trying KNNimputer with the five fold cross validation. Here is the code:

from numpy import mean
from numpy import std
from pandas import read_csv
from sklearn.ensemble import RandomForestClassifier
from sklearn.impute import SimpleImputer
from sklearn.impute import KNNImputer
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import RepeatedStratifiedKFold
from sklearn.pipeline import Pipeline
from pandas import read_csv
imputer = KNNImputer(n_neighbors=5, weights='uniform', metric='nan_euclidean')
df=read_csv('data.csv', header=None,na_values='?')
data=df.values
ix = [i for i in range(data.shape[1]) if i != 49]
X, y = data[:, ix], data[:, 49]
model = RandomForestClassifier()
pipeline = Pipeline(steps=[('i', imputer), ('m', model)])
cv = RepeatedStratifiedKFold(n_splits=5, n_repeats=1, random_state=1)
scores = cross_val_score(pipeline, X, y, scoring='accuracy', cv=cv, n_jobs=-1)

But the problem here is I don't need a score. I want the dataset (in five folds or whole) after filling the missing values in the folds because I need to do feature selection using the five folds after the imputation and then classification. So how can i get the dataset after imputation?

Answer 1

As discussed in the comments, the CV procedure will not be of any actual help here. What you actually need is:

• Fit your KNNImputer and use it to transform (impute) your training data
• Use this already fitted imputer to transform accordingly your unseen data

This way, both your training and test data will be sharing a common impute procedure, hence whatever feature selection method you choose will be actually applicable to both datasets.

Here is a demonstration with dummy data, adapting the example from the documentation:

import numpy as np
from sklearn.impute import KNNImputer

X = [[1, 2, np.nan], [3, 4, 3], [np.nan, 6, 5], [8, 8, 7]] # dummy data
imputer = KNNImputer(n_neighbors=2)
X_imp = imputer.fit_transform(X) # fit imputer & transform training dta in 1 step
X_imp
# result:
array([[1. , 2. , 4. ],
       [3. , 4. , 3. ],
       [5.5, 6. , 5. ],
       [8. , 8. , 7. ]])

# new (unseen - test) data with missing values:
# we DON'T fit the imputer again
X_new = np.array([[7, 3, 4], [np.nan, 8, 7]])
X_new_imp = imputer.transform(X_new) # use the imputer already fitted with the training data
X_new_imp
# result:
array([[7. , 3. , 4. ],
       [5.5, 8. , 7. ]])