Fast Adapting Ensemble: A New Algorithm for Mining Data Streams with Concept Drift

Algorithm 1

: Data stream.

: Each training example is formed by a vector and a discrete

value , named label and taken from a finite set named class.

: Vector of base classifiers.

: Vector of weights of base classifiers.

: Vector of status of base classifiers (Active or inactive).

: Vector of concepts associated with base classifiers.

: Ensemble.

: Set of examples necessary for building a new base classifier.

: Set of examples necessary for testing the ensemble.

ne: Number of examples necessary for building a new base classifier.

nt: Number of examples necessary for testing the ensemble.

General FAE algorithm

Initialization_ensemble

While (next  example)  // Start the training of the ensemble.

 if (  mod nt = 0)  // each nt examples weights and status are updated.

  Update_base_classifier_weight

  Update_base_classifier_status

 end if    // End of the update block

 if (  mod ne = 0) //each ne examples creating a new base classifier is analyzed.

  Add_new_base_classifier

 end if

end while