Algorithm 1 Feature selection.

Input: Original features $\mathsf{\Phi }$, input data $\{({\mathrm{X}}_1,y_1),\dots ,({\mathrm{X}}_N,y_N)\}$

Output: Selected features ${\mathsf{\Phi }}^{\prime }$

1: Initialize thresholds $r_1$, $r_2$

2: Initialize number of repeats $T$

3: for $f \mathrm{in} \mathsf{\Phi }$ do

4:     Compute actual Gini importance $I_{\left\{f\right\}}$ from $\{({\mathrm{X}}_1,y_1),\dots ,({\mathrm{X}}_N,y_N)\}$ according to Equation (1)

5: end for

6: for $i=1\to T$ do

7:     Shuffle the labels $y_1,\dots ,y_N$, which is referred to as $y_1^{\prime },\dots ,y_N^{\prime }$

8:     for $f \mathrm{in} \mathsf{\Phi }$ do

9:         Compute new Gini importance of $D_{\left\{f\right\}}^i$ from $\{({\mathrm{X}}_1,y_1^{\prime }),\dots ,({\mathrm{X}}_N,y_N^{\prime })\}$ according to Equation (1)

10:     end for

11: end for

12: for $f \mathrm{in} \mathsf{\Phi }$ do

13:     Compute score of the feature according to Equation (4), which is referred to as $S_{\left\{f\right\}}$

14: end for

15: Select the features with a score lower than $r_1$, which is referred to as ${\mathsf{\Phi }}_1$

16: Delete the features ${\mathsf{\Phi }}_1$ from $\mathsf{\Phi }$

17: Compute the correlation matrix of features $\mathsf{\Sigma }$ according to Equation (5)

18: Select the features with less actual Gini importance in each pair of features with a correlation above $r_2$ which is referred to as ${\mathsf{\Phi }}_2$

19: Delete the features ${\mathsf{\Phi }}_2$ from $\mathsf{\Phi }$

20: Obtain the remaining features in $\mathsf{\Phi }$, which is referred to as selected features ${\mathsf{\Phi }}^{\prime }$