EHN: Implementation of BalancedRandomForestClassifier (#459)

Author: glemaitre

README.rst

@@ -159,9 +159,11 @@ Below is a list of the methods currently implemented in this module.
     1. SMOTE + Tomek links [12]_
     2. SMOTE + ENN [11]_
 
-* Ensemble sampling
+* Ensemble classifier using samplers internally
     1. EasyEnsemble [13]_
     2. BalanceCascade [13]_
+    3. Balanced Random Forest [16]_
+    4. Balanced Bagging
 
 The different algorithms are presented in the sphinx-gallery_.
 
@@ -200,3 +202,5 @@ References:
 .. [14] : I. Tomek, “An experiment with the edited nearest-neighbor rule,” IEEE Transactions on Systems, Man, and Cybernetics, vol. 6(6), pp. 448-452, 1976. [`bib <references.bib#L158>`_]
 
 .. [15] : H. He, Y. Bai, E. A. Garcia, S. Li, “ADASYN: Adaptive synthetic sampling approach for imbalanced learning,” In Proceedings of the 5th IEEE International Joint Conference on Neural Networks, pp. 1322-1328, 2008. [`pdf <https://pdfs.semanticscholar.org/4823/4756b7cf798bfeb47328f7c5d597fd4838c2.pdf>`_] [`bib <references.bib#L62>`_] 
+
+.. [16] : C. Chao, A. Liaw, and L. Breiman. "Using random forest to learn imbalanced data." University of California, Berkeley 110 (2004): 1-12.

doc/api.rst

@@ -109,6 +109,7 @@ Prototype selection
 
    ensemble.BalanceCascade
    ensemble.BalancedBaggingClassifier
+   ensemble.BalancedRandomForestClassifier
    ensemble.EasyEnsemble
    ensemble.EasyEnsembleClassifier
 

doc/ensemble.rst

@@ -116,24 +116,24 @@ random under-sampler::
          [   0,   55,    4],
          [  42,   46, 1091]])
 
-It also possible to turn a balanced bagging classifier into a balanced random
-forest using a decision tree classifier and setting the parameter
-``max_features='auto'``. It allows to randomly select a subset of features for
-each tree::
-
-  >>> brf = BalancedBaggingClassifier(
-  ...          base_estimator=DecisionTreeClassifier(max_features='auto'),
-  ...          random_state=0)
+:class:`BalancedRandomForestClassifier` is another ensemble method in which
+each tree of the forest will be provided a balanced boostrap sample. This class
+provides all functionality of the
+:class:`sklearn.ensemble.RandomForestClassifier` and notably the
+`feature_importances_` attributes::
+
+
+  >>> from imblearn.ensemble import BalancedRandomForestClassifier
+  >>> brf = BalancedRandomForestClassifier(n_estimators=10, random_state=0)
   >>> brf.fit(X_train, y_train) # doctest: +ELLIPSIS
-  BalancedBaggingClassifier(...)
+  BalancedRandomForestClassifier(...)
   >>> y_pred = brf.predict(X_test)
   >>> confusion_matrix(y_test, y_pred)
   array([[   9,    1,    2],
-         [   0,   54,    5],
-         [  31,   34, 1114]])
-
-See
-:ref:`sphx_glr_auto_examples_ensemble_plot_comparison_bagging_classifier.py`.
+         [   3,   54,    2],
+         [ 113,   47, 1019]])
+  >>> brf.feature_importances_
+  array([ 0.63501243,  0.36498757])
 
 A specific method which uses ``AdaBoost`` as learners in the bagging
 classifier is called EasyEnsemble. The :class:`EasyEnsembleClassifier` allows
@@ -149,4 +149,7 @@ the ensemble as::
   >>> confusion_matrix(y_test, y_pred)
   array([[  9,   1,   2],
          [  5,  52,   2],
-         [252,  45, 882]])
+         [252,  45, 882]])
+
+See
+:ref:`sphx_glr_auto_examples_ensemble_plot_comparison_ensemble_classifier.py`.

doc/whats_new/v0.0.4.rst

@@ -33,6 +33,10 @@ New features
   AdaBoost classifier trained on balanced bootstrap samples.
   :issue:`455` by :user:`Guillaume Lemaitre <glemaitre>`.
 
+- Add :class:`imblearn.ensemble.BalancedRandomForestClassifier` which balanced
+  each bootstrap provided to each tree of the forest.
+  :issue:`459` by :user:`Guillaume Lemaitre <glemaitre>`.
+
 Enhancement
 ...........