Algorithm 2 MRMR algorithm.

Procedure MRMR($\mathbf{X}$, $\mathbf{Y}$): training dataset 01:   select (${max}_{x\in \varphi }M{A}_x$): the most relevance feature 02:   include ($x\subset S$): the selected feature x to an empty set S $\mathbf{do}$: 03:   find ($M{A}_x\ne 0,\in S^c,M{I}_x=0,\in S^c$): where $S^c$ is the complement of S       $\mathbf{if}$ ($M{A}_x\ne 0,\notin S^c$) and ($M{I}_x=0,\notin S^c$) go to line 6:       $\mathbf{else}$ 04:      select (${max}_{x\in S^c,M{I}_x=0}M{A}_x$): the most relevance feature 05:      include ($x\subset S$): the selected feature x to the set S       $\mathbf{endif}$ $\mathbf{while}$: until $M{I}_x\ne 0$ for ∀ feature $\in S^c$ $\mathbf{do}$: 06:   select (${max}_{x\in S^C}MI{Q}_x$): the feature with the nonzero relevance and redundancy 07:   include ($x\subset S$): the selected feature x to the set S $\mathbf{while}$: until $M{A}_x=0$ for ∀ feature $\in S^c$ 08:   include ($M{A}_x=0$): the feature to the set S in random order ($x\subset S$) End procedure