Algorithm 2.

HUSDOS-Boost with AdaBoost.M1

1: Initialize the boosting weights Dn,1 = 1/N for xn ∈ S, and the sampling weights $S{W}_{1,n}=1/N^{maj}$ for $x_n\in S^{maj}$.

2: for t = 1, …, T do

3:     Apply HUS with SWt,n to Smaj to generate ${Ŝ}_t^{maj}$ with a size Nu.

4:     Apply DOS to Smin to generate ${Ŝ}_t^{min}$ with a size No, where $S^{min}\subset {Ŝ}_t^{min}$.

5:     ${Ŝ}_t={Ŝ}_t^{maj}\cup {Ŝ}_t^{min}$.

6:     Train the tth weak classifier ft from Ŝt so as to minimize Ĵt.

7:     Get hypothesis of xn ∈ S: ht,n = ft(xn).

8:     Calculate the error of ht,n, εt: ${\epsilon }_t={\sum }_{n:h_{t,n}\ne y_n}{D}_{t,n}$.

9:     Set βt = εt/(1 − εt).

10:     Update the boosting weights Dt+1,n by Eq.(2).

11:     Update the sampling weights SWt+1,n by Eq.(4).

12: end for

13: return The final hypothesis H(x).