Algorithm: SVM-PSO

Inputs: nparticles, c1, c2, Wmin, Wmax, dataset, Cmin, Cmax, N

Output: gbest containing C, γ, and selected features

1

Initialize random uniform lists rc and rγwith the size of nparticles and random uniform matrix Rfeatures with Size nfeatures × nparticles

2

Initialize matrix X by vertically stacking rC, rγ, and Rfeatures and matrix V with the size of (nfeatures+ 2) × nparlicles

3

Pbest ← X

4

gbest ← column of pbest with lowest cross-entropy loss of SVM

5

Create w as a list of evenly spaced numbers (in descending order) from Wmin to Wmax

6

fori = 1,2,…,Ndo

7

  Wcurrent←Wi

8

  Create random variables r1 and r2 on a continuous uniform distribution

9

  V←Wcurrent V+c1r1(pbest - X) + c2r2(gbest−X)

10

  X←X+V

11

  Replace the first row’s negative values of X with random uniform numbers betweenCmin and Cmax

12

  Replace the second row’s negative values of X with random uniform numbers between 0 and 1

13

  Run SVM with each particle’s C, γ, and selected features

14

  Calculate cross-entropy as the loss Junction of each SVM and store the list of results in objectiveresult

15

  Replace pbest of particles that have better objectiveresult

16

  Update gbest if necessary

17

end for

18

Return gbest

Pseudocode 1: A standard SVM-PSO.