Adaptive Ensemble with Human Memorizing Characteristics for Data Stream Mining

Algorithm 1

Memorizing based Adaptive Ensemble.

Input: : data stream of instances

    : memory capacity, the maximum number of component classifiers in the knowledge repository.

    : the maximum number of classifiers that can be recalled

() Initialization: ; ; ;

() for all  data chunks   do

   (2.1) new component classifier built on ;

   (2.2) Initialize parameters for classifier :

     ; ; ; ;

   (2.3) Add to the knowledge repository:

     ;

   (2.4) ES = ensemble-pruning ;

   (2.5) for all  classifiers   do

    (2.5.1) ;

    (2.5.2) ;

    (2.5.3) compute the forgetting factor of based on (13);

   (2.6) end for

   (2.7) for all  classifiers   do

    (2.7.1) update the memory retention value of based on (12);

   (2.8) end for

   (2.9) if    then  remove the classifier with the lowest memory retention value from ;

   (2.10) ;

() end for