Algorithm 3. The training phase of the PCC-FS algorithm

Input: original training data, iterative times T Output: $T\ast Q$ binary classifiers ${\mathrm{ƕ}}_{r,j},$ $r=1,2,\dots ,T;j=1,2,\dots ,Q$ Steps generate ${\mathfrak{X}}_{\mathit{t}\mathit{r}\mathit{a}\mathit{i}\mathit{n}}$ through feature selection process; learn label couplings and generate ${\mathrm{Ɣ}}_j$ for each label by Equation (14), $j=1,2,\dots ,Q$; initialize ${\mathit{\chi }}_{\mathit{t}\mathit{r}\mathit{a}\mathit{i}\mathit{n}}^{\mathbf{1},\mathit{j}}$ with zero matrix ${\left[\mathbf{0}\right]}_{\mathit{m}\times \mathit{Q}}$; for $r\in 1,2,\dots ,T$; for $j\in 1,2,\dots ,Q$; $\mathit{l}\mathit{p}\mathit{r}{\mathit{e}}^{\mathit{r},\mathit{j}}=\mathrm{Ƈ}\left({\mathrm{Ɣ}}_j\right)$; ${\mathit{x}}^{\mathit{r},\mathit{j}}=\left[{\mathfrak{X}}_{\mathit{t}\mathit{r}\mathit{a}\mathit{i}\mathit{n}},\mathit{l}\mathit{p}\mathit{r}{\mathit{e}}^{\mathit{r},\mathit{j}}\right]$; ${\mathit{\chi }}_{\mathit{t}\mathit{r}\mathit{a}\mathit{i}\mathit{n}}^{\mathit{r},\mathit{j}}=\left[{\mathit{x}}^{\mathit{r},\mathit{j}},{\mathfrak{Y}}_{\mathit{t}\mathit{r}\mathit{a}\mathit{i}\mathit{n}}\left(\mathit{j}\right)\right]$; ${\mathrm{ƕ}}_{r,j}\leftarrow B({\mathit{\chi }}_{\mathit{t}\mathit{r}\mathit{a}\mathit{i}\mathit{n}}^{\mathit{r},\mathit{j}}$); $\mathrm{Ƈ}\left(\mathit{j}\right)={\mathrm{ƕ}}_{r,j}\left({\mathit{x}}^{\mathit{r},\mathit{j}}\right)$; end for; end for.