Algorithm 1: Reduction of features.

Require: d × N feature matrix F.

1: Estimate effective-rank k of $\mathbf{F}$ (using rank estimation technique in [40]).

2: Find a sub matrix with k rows and call it ${\mathbf{F}}_k$.

3: while effective-$rank\left({\mathbf{F}}_k\right)\ne k$ do

4:     Find another sub matrix with k rows and call it ${\mathbf{F}}_k$.

5:  end while

6: k features corresponding to each row of ${\mathbf{F}}_k$ are most relevant features.