Algorithm 3 Fitness Function

Input:$N$: The samples amount of training set. $K$: The amount of feature subsets in an individual. The number of weak classifiers to train. $I$: An individual. $T$: Training Set. Output: $V$: Fitness value of $I$. 1:    For $k=1,\dots ,K$ do 2:    Divide features into Binary Feature $FB$ and Continuous Feature $FC$. 3:    Rotation: Apply $PCA$ to $FC$ and apply $MCA$ to $FB$ and then merge the two parts. 4:    Initialize the weights $w^1=\left[w_1,\dots ,w_N\right], w_j^1=\frac{1}{N}.{{\displaystyle \sum }}_{j=1}^N{w}_j^1=1.$ 5:    For $k=1,\dots ,K$ do 6:    Take a sample $S_k$ from $T$ using distribution $w^k$. 7:    Train a classifier $D_k$ using $S_k$ as the training set. 8:    Calculate the weighted ensemble error at step $k$ by ${ϵ}_k={\displaystyle \sum }_{j=1}^N{w}_j^k{l}_k^j,$ ($l_k^j=1$ if $D_k$ misclassifies $t_j$ and $l_k^j=0$ otherwise.) 9:    If ${ϵ}_k=0 or {ϵ}_k\ge 0.5$, ignore $D_k$, reinitialize the weights $w_j^k$ to $\frac{1}{N}$ and continue. 10:     Else, calculate ${\beta }_k=\frac{{ϵ}_k}{1-{ϵ}_k},$ where ${ϵ}_k\in \left(0,0.5\right)$. 11:     Update the $k^{th}$ part weights in $I$ by $w_j^{k+1}=\frac{w_j^k{\beta }^{\left(1-l_k^j\right)}}{{{\displaystyle \sum }}_{i=1}^N{w}_i^k{\beta }^{\left(1-l_k^i\right)}}, j=1,\dots ,N.$ 12:    Calculate the support for each class $w_t$ in Validation Set by ${\mu }_t\left(x\right)={\displaystyle \sum }_{D_k\left(x\right)=w_t}\ln \left(\frac{1}{{\beta }_k}\right).$ 13:     The class with the maximum support is chosen as the label for $x$. 14:     $V$ is calculated by $F1-score=\frac{2\times precision\times recall}{precision+recall}$n validation data.