Constructing Better Classifier Ensemble Based on Weighted Accuracy and Diversity Measure

Pseudocode 1

The pseudocode to compute the WAD score.

Input:

 (1) Original ensemble with m classifiers: 𝐸={𝐶₁,𝐶₂,…,𝐶_𝑚}

 (2) Validation dataset: 𝐷={𝑆₁,𝑆₂,…,𝑆_𝑘}

Define:

 (1) Classifier predictions: Preds

 (2) Ensemble accuracy: Acc

 (3) Ensemble diversity: Div

 (4) The weight parameters: 𝛼 and 𝛽

Output:

 (1) WAD: ensemble quality score

Begin

 (1) For each 𝐶_𝑖 in ensemble E:

   for each 𝑆_𝑗 invalidation dataset D:

    get the jth classifier 𝐶_𝑖’s prediction ℎ_𝑖(𝑥_𝑗) on 𝑆_𝑗 and put it in Preds

   end for

  End For

 (2) Compute accuracy Acc of E according to Notation 2

 (3) Compute diversity Div of E according to Notation 4

 (4) Estimate 𝛼 and 𝛽 according to Lemma 2

 (5) Compute the score by

      WAD=Acc⋅Div𝛽⋅Acc+𝛼⋅Div

End;