Algorithm 1.

HMCBCG model

Input: Train data Dtr, maximum cluster numbers kmax, m base learners L1, L2,…,Lm. Set the number of samples in each subset obtained by bagging to n.

Output: Candidate classifier pool Ψ

1: Ψ←∅

2: for 2 to kmaxdo:

3: Use k-means to divide Dtr into k clusters.

4: for each cluster do:

5: Apply GA to optimize SVM to get the SVM with optimal parameters.

6: Add the trained SVM to Ψ.

7: end for 8: Shuffle all the clusters to restore Dtr.

9: end for

10: for 1 to m do:

11: for 1 to n do:

12: Randomly draw a sample from Dtr.

13: end for

14: Use the n samples to train a base learner.

15: Add the trained base learner to Ψ. 16: Put the n samples back to restore Dtr.

17: end for

18: return Ψ