Get OOB samples of random forest in sklearn [closed]

Question

In sklearn I oob_score_ gives me the OOB score of a random forest model. This score is calculated by the samples which were left out during RF training. Is there a way to get the individual OOB samples to analyse which samples were predicted correctly or not?

Answer 1

If you want to know the OOB prediction, it is stored under the attributes oob_decision_function_ . This is a probability obtained by averaging predictions across all your trees where the row or observation is OOB.

First use an example dataset:

import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import make_classification
from sklearn.metrics import accuracy_score

X, y = make_classification(n_samples=1000, n_features=4,
n_informative=2, n_redundant=0,random_state=123, shuffle=False)

clf = RandomForestClassifier(max_depth=2, random_state=123,oob_score=True)
clf.fit(X,y)

clf.oob_score_
0.95

Then we can check the OOB predictions:

clf.oob_decision_function_
Out[115]: 
array([[0.76976691, 0.23023309],
       [0.78910912, 0.21089088],
       [0.398826  , 0.601174  ],
       ...,
       [0.1952597 , 0.8047403 ],
       [0.12296672, 0.87703328],
       [0.13481899, 0.86518101]])

You can get the predicted labels:

pred = np.argmax(clf.oob_decision_function_,axis=1)

Check against the accuracy:

accuracy_score(pred,y)
0.95