Algorithm 1.

Greedy Feature Selection

Input: Input Feature Matrix F containing N samples of feature vector ${\left\{f_i\right\}}_{i=1}^N$

Output: Selected Feature List l

1:  Step 0: Initialization

2:  Put ith feature with the best accuracy into list l

3:  l ← argmaxi P(fi)

4:  Initialize the best accuracy B ← 0

5:  Initialize local best accuracy LB ← P(fl)

6:  Delete fi

7:

8:  Step 1: Greedy Feature Selection

9:  while LB > B do

10:      B ← LB

11:      l ← argmaxi P(< fi, fl >)

12:      LB ← P(fl)

13:      Delete fi

14: return l