Begin

Initialize particle swarm, including the settlement of the original position X_i and original speed V_i.

Repeat

Transfer the form of the parameters of LS-SVM, according to the binary form of the grains.

Compare the value of the fitness function of every grain f(x_i) with its optimum value f(pbest_i).

If

f(x_i) < f(pbest_i)

Make the value of the fitness function take place of the value of the fitness function before, and so it is with the grain.

Compare the value of the fitness function of every grain f(x_i) with the global optimum value f(gbest_i).

If

f(x_i) < f(gbest_i),

Make the value of the fitness function of that grain take place of the optimum value of the fitness function before, and so it is with the grain.

Update the speed and value of the grains according to the Equations (1) and (2), and build the new population X_i+1. Update the speed according to the rules: if v_i  > V_max, v_i = V_max, if v_i ≤ −V_max, v_i = −V_max.

Update the inertia factor w.

Update the binary digit of the grains.

Check whether the requirement of ending is satisfied. The requirement is reaching the max iterations or the evaluation of estimate is less than the given accuracy.

Improve the iteration. Make T = T + 1.

Until

The requirement of ending is satisfied, end the selection for the best feature, and return the best character subset and classification accuracy.