Improving Classification Performance through an Advanced Ensemble Based Heterogeneous Extreme Learning Machines

Algorithm 1

AELME.

Training phase ( )

Input:  Original training dataset # of hidden nodes

 # of iterations and # of parts

Output:  Ensemble classifier model

(1) Split the original training dataset:  

   Initialization

(2) for    do

(3)      Set Reconstruct training data by re-sampling on

(4)      Random Select a member of ELM () type out of three types .

(5)      Train a member of ELM ( on )

(6)      Test the selected member of ELM ( on )

(7)      Add classifier to the ensemble

(8) AccOld = Accuracy of

(9) DivOld = Diversity of

(10)

(11) for    do

(12)      for    do

(13)    Set Reconstruct training data by re-sampling on

(14)    Random Select a member of ELM () type out of three types .

(15)    Train a member of ELM ( on )

(16)    Test the selected member of ELM ( on )

(17)    Add classifier to the ensemble

(18)      Add to the Ensemble

(19)      AccNew = Accuracy of

(20)      DivNew = Diversity of

(21)      if ((AccNew AccOld) and (DivNew DivOld)) then

(22)    AccOld = AccNew

(23)    DivOld = DivNew

(24)      else

(25)    Exclude from

Prediction phase( )

Input: Unknown sample , ensembles classifier model:

Output: Class label of sample .

(26) Loop for

(27) Vote on all the outputs , then output the class label of with the highest votes.