A Weighted Voting Classifier Based on Differential Evolution

Algorithm 1

DE algorithm for DEWVote’s model selection.

Input: The control parameters of DE: mutation factor F, crossover rate CR, and population size N.

(1)       Initialization(); {Generate uniformly distributed random population of N individuals

      𝑋_𝐺=𝑋_1,𝐺,𝑋_2,𝐺,…,𝑋_𝑖,𝐺,…,𝑋_𝑁,𝐺, where 𝑋_𝑖,𝐺=𝑥_𝑖,𝐺(1),𝑥_𝑖,𝐺(2),…,𝑥_𝑖,𝐺𝑗,…,𝑥_𝑖,𝐺(𝐷) is a

       vector representing the weights (𝑤₁, 𝑤₂, …, 𝑤_𝑗,…, 𝑤_𝐷) of D base classifiers.}

(2)      Set the generation iterator 𝐺=0.

(3)        while the stopping criterion is not satisfied do

(4)     for  (𝑖=0;𝑖<𝑁;𝑖++)   do

(5)     Select random indexes𝑟₁, 𝑟₂, and 𝑟₃ to be different from each other and from the index i.

(6)     Compute a mutant vector 𝑉_𝑖,𝐺 using (1).

(7)     Generate random number 𝑗_rand.

(8)     for  𝑗=0;𝑗<𝐷;𝑗++   do

(9)        Decide trial individual 𝑇_𝑖,𝐺 using (6).

(10)      end for

(11)         Compute the fitness of the vector 𝑇_𝑖,𝐺 and 𝑋_𝑖,𝐺 using 10-fold cross validation, and

        update the vector 𝑋_𝑖,𝐺+1 of the next generation (𝐺+1) using (7).

(12)    end for

(13)    Update generation iterator 𝐺=𝐺+1.

(14) end  while

 Output: The optimal weights (𝑤₁, 𝑤₂, …, 𝑤_𝑗, …, 𝑤_𝐷) for DEWVote.