Algorithm 2 M-SVM-RFE-OA

Input: training dataset X, t.

Output: selected feature subset FS.

Begin

c_acc = 0;   c_oa = ∞;

F = {all input features};

While (|F| > 0) Do

Calculate Nr(x) for each x ∈ X based on F;

Xt = X;

For each c ∈ C Do

Xc = {x | x ∈ X, Label(x) = c and Nr(x) > 0};

θ = |{x | x ∈ X, Label(x) = c}|;

If |Xc| > θ/3 Then

Rank the samples in Xc based on Nr(x) in descending order;

Xt = Xt − {θ/3 top ranked samples in Xc};

Else

Xt = Xt − Xc;

End if;

End for;

Construct an SVM based on Xt and F;

T_c_acc = accuracy rate of SVM;

T_c_oa = average Nr(x) of the samples in Xt based on F;

If   T_c_acc − T_c_oa > c_acc − c_oa Then

c_acc = T_c_acc;

c_oa = T_c_oa;

FS = F;

End if;

Rank the features in F by |w| in descending order;

F = F − {t × |F| bottom ranked features in F};

End while;

Return FS;

End