Algorithm 1 The pseudo-code of the algorithm to convert the cost matrix to multi-label classification cost. C′ denotes the cost matrix specifying the misclassification cost between each pair of categories. Y is the label matrix of the training dataset. c′ contains the derived false positive costs for each category

$\mathit{c}\prime$= computeFalsePositiveCosts ($C\prime$,$Y$)

# Normalize the training set labels $Y$ to $Y^n$ for i $=$ 1 to N do # Calculate the label number of samples i $L\left[i\right]=$ max(sum($Y\left[i,1:m\right]$), 1) # Normalize the training set labels $Y$ $Y^n\left[i, 1:m\left]=Y\right[i, 1:m\right]/L\left[\mathrm{i}\right]$ endfor # Calculate the misclassification cost of each label on each sample $S{C}^{\prime }=$ matrix_multiply ($Y^n$, $C\prime$) # Mask out the entries with positive labels $S{C}^{″}=S{C}^{\prime }\cdot \left(1-Y\right)$ # Calculate the mean false positive costs of each label in the training set for i $=$1 to m do ${\mathit{c}}^{\prime }\left[\mathrm{i}\right]$ $=$ $\mathrm{sum}\left(S{C}^{″}\left[1:N, i\right]\right)/\mathrm{sum}\left(1-Y\left[1:N,i\right]\right)$ endfor Return ${\mathit{c}}^{\prime }$